From e6f6aad86a5a12a1c040b47a520b976ff79dd510 Mon Sep 17 00:00:00 2001 From: James Ketrenos Date: Thu, 24 Apr 2025 16:08:26 -0700 Subject: [PATCH] Working with refactored code --- frontend/src/App.css | 24 +- frontend/src/App.tsx | 54 +-- frontend/src/ChatBubble.tsx | 27 +- frontend/src/Controls.tsx | 4 +- frontend/src/Conversation.tsx | 311 +++++++++------ frontend/src/DocumentViewer.tsx | 625 ++++++++++++++---------------- frontend/src/Message.tsx | 29 +- frontend/src/ResumeBuilder.tsx | 16 +- frontend/src/Snack.tsx | 4 +- frontend/src/VectorVisualizer.tsx | 1 - src/.gitignore | 2 + src/server.py | 414 ++++++++++++++------ src/utils/rag.py | 12 +- 13 files changed, 896 insertions(+), 627 deletions(-) create mode 100644 src/.gitignore diff --git a/frontend/src/App.css b/frontend/src/App.css index cce083c..a034a7c 100644 --- a/frontend/src/App.css +++ b/frontend/src/App.css @@ -62,14 +62,6 @@ button { justify-self: end; /* Align the first column content to the right */ } -.ChatBox { - display: flex; - flex-direction: column; - flex-grow: 1; - max-width: 1024px; - margin: 0 auto; -} - .DocBox { display: flex; flex-direction: column; @@ -112,17 +104,27 @@ button { padding-right: 16px !important; } +.ChatBox { + display: flex; + flex-direction: column; + flex-grow: 1; + max-width: 1024px; + width: 100%; + margin: 0 auto; + background-color: #D3CDBF; +} + .Conversation { display: flex; background-color: #F5F5F5; border: 1px solid #E0E0E0; flex-grow: 1; - overflow-y: auto; padding: 10px; flex-direction: column; - height: 100%; - max-height: 100%; font-size: 0.9rem; + width: 100%; + /* max-width: 1024px; */ + margin: 0 auto; } .user-message.MuiCard-root { diff --git a/frontend/src/App.tsx b/frontend/src/App.tsx index f8500db..f2384f0 100644 --- a/frontend/src/App.tsx +++ b/frontend/src/App.tsx @@ -20,7 +20,7 @@ import MenuIcon from '@mui/icons-material/Menu'; import { ResumeBuilder } from './ResumeBuilder'; import { Message, ChatQuery, MessageList, MessageData } from './Message'; -import { SeverityType } from './Snack'; +import { SetSnackType, SeverityType } from './Snack'; import { VectorVisualizer } from './VectorVisualizer'; import { Controls } from './Controls'; import { Conversation, ConversationHandle } from './Conversation'; @@ -68,7 +68,6 @@ function CustomTabPanel(props: TabPanelProps) { } const App = () => { - const conversationRef = useRef(null); const [processing, setProcessing] = useState(false); const [sessionId, setSessionId] = useState(undefined); const [connectionBase,] = useState(getConnectionBase(window.location)) @@ -86,11 +85,13 @@ const App = () => { const chatRef = useRef(null); // Set the snack pop-up and open it - const setSnack = useCallback((message: string, severity: SeverityType = "success") => { - setSnackMessage(message); - setSnackSeverity(severity); - setSnackOpen(true); - }, []); + const setSnack: SetSnackType = useCallback((message: string, severity: SeverityType = "success") => { + setTimeout(() => { + setSnackMessage(message); + setSnackSeverity(severity); + setSnackOpen(true); + }); + }, [setSnackMessage, setSnackSeverity, setSnackOpen]); useEffect(() => { if (prevIsDesktopRef.current === isDesktop) @@ -172,27 +173,34 @@ What would you like to know about James? const url = new URL(window.location.href); const pathParts = url.pathname.split('/').filter(Boolean); + const fetchSession = async () => { + try { + const response = await fetch(connectionBase + `/api/context`, { + method: 'POST', + headers: { + 'Content-Type': 'application/json', + }, + }); + + if (!response.ok) { + throw Error("Server is temporarily down."); + } + const data = await response.json(); + setSessionId(data.id); + window.history.replaceState({}, '', `/${data.id}`); + } catch (error: any) { + setSnack("Server is temporarily down", "error"); + }; + }; + if (!pathParts.length) { console.log("No session id -- creating a new session") - fetch(connectionBase + `/api/context`, { - method: 'POST', - headers: { - 'Content-Type': 'application/json', - }, - }) - .then(response => response.json()) - .then(data => { - console.log(`Session id: ${data.id} -- returned from server`) - setSessionId(data.id); - window.history.replaceState({}, '', `/${data.id}`); - }) - .catch(error => console.error('Error generating session ID:', error)); + fetchSession(); } else { console.log(`Session id: ${pathParts[0]} -- existing session`) setSessionId(pathParts[0]); } - - }, [setSessionId, connectionBase]); + }, [setSessionId, connectionBase, setSnack]); const handleMenuClose = () => { setIsMenuClosing(true); @@ -395,7 +403,7 @@ What would you like to know about James? - + void, + setSnack: SetSnackType, }; type GPUInfo = { diff --git a/frontend/src/Conversation.tsx b/frontend/src/Conversation.tsx index 7afb54f..a8bcf90 100644 --- a/frontend/src/Conversation.tsx +++ b/frontend/src/Conversation.tsx @@ -11,10 +11,10 @@ import { SxProps, Theme } from '@mui/material'; import PropagateLoader from "react-spinners/PropagateLoader"; import { Message, MessageList, MessageData } from './Message'; -import { SeverityType } from './Snack'; +import { SetSnackType } from './Snack'; import { ContextStatus } from './ContextStatus'; -const loadingMessage: MessageData = { "role": "assistant", "content": "Establishing connection with server..." }; +const loadingMessage: MessageData = { "role": "status", "content": "Establishing connection with server..." }; type ConversationMode = 'chat' | 'job_description' | 'resume' | 'fact_check'; @@ -23,42 +23,73 @@ interface ConversationHandle { } interface ConversationProps { - className?: string, - type: ConversationMode - prompt: string, - actionLabel?: string, - resetAction?: () => void, - resetLabel?: string, - connectionBase: string, - sessionId?: string, - setSnack: (message: string, severity: SeverityType) => void, - defaultPrompts?: React.ReactElement[], - preamble?: MessageList, - hideDefaultPrompts?: boolean, - messageFilter?: (messages: MessageList) => MessageList, - messages?: MessageList, + className?: string, // Override default className + type: ConversationMode, // Type of Conversation chat + prompt?: string, // Prompt to display in TextField input + actionLabel?: string, // Label to put on the primary button + resetAction?: () => void, // Callback when Reset is pressed + multiline?: boolean, // Render TextField as multiline or not + resetLabel?: string, // Label to put on Reset button + connectionBase: string, // Base URL for fetch() calls + sessionId?: string, // Session ID for fetch() calls + setSnack: SetSnackType, // Callback to display snack popups + defaultPrompts?: React.ReactElement[], // Set of Elements to display after the TextField + defaultQuery?: string, // Default text to populate the TextField input + preamble?: MessageList, // Messages to display at start of Conversation until Action has been invoked + hidePreamble?: boolean, // Whether to hide the preamble after an Action has been invoked + hideDefaultPrompts?: boolean, // Whether to hide the defaultPrompts after an Action has been invoked + messageFilter?: ((messages: MessageList) => MessageList) | undefined, // Filter callback to determine which Messages to display in Conversation + messages?: MessageList, // sx?: SxProps, + onResponse?: ((message: MessageData) => MessageData) | undefined, // Event called when a query completes (provides messages) }; -const Conversation = forwardRef(({ ...props }: ConversationProps, ref) => { +const Conversation = forwardRef(({ + className, + type, + prompt, + actionLabel, + resetAction, + multiline, + resetLabel, + connectionBase, + sessionId, + setSnack, + defaultPrompts, + hideDefaultPrompts, + defaultQuery, + preamble, + hidePreamble, + messageFilter, + messages, + sx, + onResponse +}: ConversationProps, ref) => { const [query, setQuery] = useState(""); const [contextUsedPercentage, setContextUsedPercentage] = useState(0); const [processing, setProcessing] = useState(false); const [countdown, setCountdown] = useState(0); const [conversation, setConversation] = useState([]); + const [filteredConversation, setFilteredConversation] = useState([]); + const [processingMessage, setProcessingMessage] = useState(undefined); const timerRef = useRef(null); const [lastEvalTPS, setLastEvalTPS] = useState(35); const [lastPromptTPS, setLastPromptTPS] = useState(430); const [contextStatus, setContextStatus] = useState({ context_used: 0, max_context: 0 }); const [contextWarningShown, setContextWarningShown] = useState(false); const [noInteractions, setNoInteractions] = useState(true); - const setSnack = props.setSnack; + const conversationRef = useRef([]); + + // Keep the ref updated whenever items changes + useEffect(() => { + conversationRef.current = conversation; + }, [conversation]); // Update the context status const updateContextStatus = useCallback(() => { const fetchContextStatus = async () => { try { - const response = await fetch(props.connectionBase + `/api/context-status/${props.sessionId}/${props.type}`, { + const response = await fetch(connectionBase + `/api/context-status/${sessionId}/${type}`, { method: 'GET', headers: { 'Content-Type': 'application/json', @@ -78,51 +109,75 @@ const Conversation = forwardRef(({ ...pro } }; fetchContextStatus(); - }, [setContextStatus, props.connectionBase, setSnack, props.sessionId, props.type]); + }, [setContextStatus, connectionBase, setSnack, sessionId, type]); + + /* Transform the 'Conversation' by filtering via callback, then adding + * preamble and messages based on whether the conversation + * has any elements yet */ + useEffect(() => { + let filtered = []; + if (messageFilter === undefined) { + filtered = conversation; + } else { + //console.log('Filtering conversation...') + filtered = messageFilter(conversation); /* Do not copy conversation or useEffect will loop forever */ + //console.log(`${conversation.length - filtered.length} messages filtered out.`); + } + if (filtered.length === 0) { + setFilteredConversation([ + ...(preamble || []), + ...(messages || []), + ]); + } else { + setFilteredConversation([ + ...(hidePreamble ? [] : (preamble || [])), + ...(messages || []), + ...filtered, + ]); + }; + }, [conversation, setFilteredConversation, messageFilter, preamble, messages, hidePreamble]); // Set the initial chat history to "loading" or the welcome message if loaded. useEffect(() => { - if (props.sessionId === undefined) { - setConversation([loadingMessage]); + if (sessionId === undefined) { + setProcessingMessage(loadingMessage); return; } const fetchHistory = async () => { try { - const response = await fetch(props.connectionBase + `/api/history/${props.sessionId}/${props.type}`, { + const response = await fetch(connectionBase + `/api/history/${sessionId}/${type}`, { method: 'GET', headers: { 'Content-Type': 'application/json', }, }); + if (!response.ok) { throw new Error(`Server responded with ${response.status}: ${response.statusText}`); } + const data = await response.json(); - console.log(`History returned from server with ${data.length} entries`) + + console.log(`History returned for ${type} from server with ${data.length} entries`) if (data.length === 0) { - setConversation([ - ...(props.preamble || []), - ...(props.messages || []), - ]); + setConversation([]) setNoInteractions(true); } else { - setConversation([ - ...(props.messages || []), - ...(props.messageFilter ? props.messageFilter(data) : data) - ]); + setConversation(data); setNoInteractions(false); } + setProcessingMessage(undefined); updateContextStatus(); } catch (error) { console.error('Error generating session ID:', error); + setProcessingMessage({ role: "error", content: "Unable to obtain history from server." }); setSnack("Unable to obtain chat history.", "error"); } }; - if (props.sessionId !== undefined) { - fetchHistory(); - } - }, [props.sessionId, setConversation, updateContextStatus, props.connectionBase, setSnack, props.preamble, props.type]); + + fetchHistory(); + }, [setConversation, setFilteredConversation, updateContextStatus, connectionBase, setSnack, type, sessionId]); const isScrolledToBottom = useCallback(()=> { // Current vertical scroll position @@ -174,12 +229,8 @@ const Conversation = forwardRef(({ ...pro }; const handleKeyPress = (event: any) => { - if (event.key === 'Enter') { - switch (event.target.id) { - case 'QueryInput': - sendQuery(query); - break; - } + if (event.key === 'Enter' && !event.shiftKey) { + sendQuery(query); } }; @@ -189,10 +240,6 @@ const Conversation = forwardRef(({ ...pro } })); - const submitQuery = (query: string) => { - sendQuery(query); - } - // If context status changes, show a warning if necessary. If it drops // back below the threshold, clear the warning trigger useEffect(() => { @@ -209,7 +256,7 @@ const Conversation = forwardRef(({ ...pro const reset = async () => { try { - const response = await fetch(props.connectionBase + `/api/reset/${props.sessionId}/${props.type}`, { + const response = await fetch(connectionBase + `/api/reset/${sessionId}/${type}`, { method: 'PUT', headers: { 'Content-Type': 'application/json', @@ -226,13 +273,7 @@ const Conversation = forwardRef(({ ...pro throw new Error('Response body is null'); } - props.messageFilter && props.messageFilter([]); - - setConversation([ - ...(props.preamble || []), - ...(props.messages || []), - ]); - + setConversation([]) setNoInteractions(true); } catch (e) { @@ -242,24 +283,40 @@ const Conversation = forwardRef(({ ...pro }; const sendQuery = async (query: string) => { + query = query.trim(); + + // If the query was empty, a default query was provided, + // and there is no prompt for the user, send the default query. + if (!query && defaultQuery && !prompt) { + query = defaultQuery.trim(); + } + + // If the query is empty, and a prompt was provided, do not + // send an empty query. + if (!query && prompt) { + return; + } + setNoInteractions(false); - if (!query.trim()) return; + if (query) { + setConversation([ + ...conversationRef.current, + { + role: 'user', + origin: type, + content: query + } + ]); + } - //setTab(0); - - const userMessage: MessageData[] = [{ role: 'user', content: query }]; + // Add a small delay to ensure React has time to update the UI + await new Promise(resolve => setTimeout(resolve, 0)); + console.log(conversation); let scrolledToBottom; - // Add user message to conversation - const newConversation: MessageList = [ - ...conversation, - ...userMessage - ]; - setConversation(newConversation); scrollToBottom(); - // Clear input setQuery(''); @@ -270,16 +327,19 @@ const Conversation = forwardRef(({ ...pro const processingId = Date.now().toString(); // Add initial processing message - setConversation(prev => [ - ...prev, - { role: 'assistant', content: 'Processing request...', id: processingId, isProcessing: true } - ]); + setProcessingMessage( + { role: 'status', content: 'Submitting request...', id: processingId, isProcessing: true } + ); + + // Add a small delay to ensure React has time to update the UI + await new Promise(resolve => setTimeout(resolve, 0)); + if (scrolledToBottom) { setTimeout(() => { scrollToBottom() }, 50); } // Make the fetch request with proper headers - const response = await fetch(props.connectionBase + `/api/chat/${props.sessionId}/${props.type}`, { + const response = await fetch(connectionBase + `/api/chat/${sessionId}/${type}`, { method: 'POST', headers: { 'Content-Type': 'application/json', @@ -334,41 +394,43 @@ const Conversation = forwardRef(({ ...pro if (update.status === 'processing') { scrolledToBottom = isScrolledToBottom(); // Update processing message with immediate re-render - setConversation(prev => prev.map(msg => - msg.id === processingId - ? { ...msg, content: update.message } - : msg - )); + setProcessingMessage({ role: 'status', content: update.message }); + // Add a small delay to ensure React has time to update the UI + await new Promise(resolve => setTimeout(resolve, 0)); if (scrolledToBottom) { setTimeout(() => { scrollToBottom() }, 50); } - - // Add a small delay to ensure React has time to update the UI - await new Promise(resolve => setTimeout(resolve, 0)); - } else if (update.status === 'done') { // Replace processing message with final result scrolledToBottom = isScrolledToBottom(); - setConversation(prev => [ - ...prev.filter(msg => msg.id !== processingId), + if (onResponse) { + update.message = onResponse(update.message); + } + setProcessingMessage(undefined); + setConversation([ + ...conversationRef.current, update.message - ]); + ]) + // Add a small delay to ensure React has time to update the UI + await new Promise(resolve => setTimeout(resolve, 0)); + const metadata = update.message.metadata; - const evalTPS = metadata.eval_count * 10 ** 9 / metadata.eval_duration; - const promptTPS = metadata.prompt_eval_count * 10 ** 9 / metadata.prompt_eval_duration; - setLastEvalTPS(evalTPS ? evalTPS : 35); - setLastPromptTPS(promptTPS ? promptTPS : 35); - updateContextStatus(); + if (metadata) { + const evalTPS = metadata.eval_count * 10 ** 9 / metadata.eval_duration; + const promptTPS = metadata.prompt_eval_count * 10 ** 9 / metadata.prompt_eval_duration; + setLastEvalTPS(evalTPS ? evalTPS : 35); + setLastPromptTPS(promptTPS ? promptTPS : 35); + updateContextStatus(); + } if (scrolledToBottom) { setTimeout(() => { scrollToBottom() }, 50); } } else if (update.status === 'error') { // Show error scrolledToBottom = isScrolledToBottom(); - setConversation(prev => [ - ...prev.filter(msg => msg.id !== processingId), - { role: 'assistant', type: 'error', content: update.message } - ]); + setProcessingMessage({ role: 'error', content: update.message }); + // Add a small delay to ensure React has time to update the UI + await new Promise(resolve => setTimeout(resolve, 0)); if (scrolledToBottom) { setTimeout(() => { scrollToBottom() }, 50); } @@ -387,8 +449,12 @@ const Conversation = forwardRef(({ ...pro if (update.status === 'done') { scrolledToBottom = isScrolledToBottom(); - setConversation(prev => [ - ...prev.filter(msg => msg.id !== processingId), + if (onResponse) { + update.message = onResponse(update.message); + } + setProcessingMessage(undefined); + setConversation([ + ...conversationRef.current, update.message ]); if (scrolledToBottom) { @@ -410,31 +476,37 @@ const Conversation = forwardRef(({ ...pro console.error('Fetch error:', error); setSnack("Unable to process query", "error"); scrolledToBottom = isScrolledToBottom(); - setConversation(prev => [ - ...prev.filter(msg => !msg.isProcessing), - { role: 'assistant', type: 'error', content: `Error: ${error}` } - ]); + setProcessingMessage({ role: 'error', content: "Unable to process query" }); setProcessing(false); stopCountdown(); if (scrolledToBottom) { setTimeout(() => { scrollToBottom() }, 50); } + // Add a small delay to ensure React has time to update the UI + await new Promise(resolve => setTimeout(resolve, 0)); } }; return ( - + { - conversation.map((message, index) => - + filteredConversation.map((message, index) => + ) } + { + processingMessage !== undefined && + + } (({ ...pro >Estimated response time: {countdown}s )} - - + {prompt && + setQuery(e.target.value)} onKeyDown={handleKeyPress} - placeholder={props.prompt} + placeholder={prompt} id="QueryInput" /> + } + { reset(); }} > - + - - + + + + - {(noInteractions || !props.hideDefaultPrompts) && props.defaultPrompts !== undefined && props.defaultPrompts.length && + {(noInteractions || !hideDefaultPrompts) && defaultPrompts !== undefined && defaultPrompts.length && { - props.defaultPrompts.map((element, index) => { + defaultPrompts.map((element, index) => { return ({element}); }) } diff --git a/frontend/src/DocumentViewer.tsx b/frontend/src/DocumentViewer.tsx index dd38ad6..b2a659f 100644 --- a/frontend/src/DocumentViewer.tsx +++ b/frontend/src/DocumentViewer.tsx @@ -1,8 +1,5 @@ -import React, { useEffect, useState, useCallback } from 'react'; +import React, { useState, useCallback, useRef } from 'react'; import { - Typography, - Card, - Button, Tabs, Tab, Paper, @@ -12,26 +9,18 @@ import { Divider, Slider, Stack, - TextField, - Tooltip } from '@mui/material'; import { useTheme } from '@mui/material/styles'; -import SendIcon from '@mui/icons-material/Send'; import { ChevronLeft, ChevronRight, SwapHoriz, - RestartAlt as ResetIcon, } from '@mui/icons-material'; -import PropagateLoader from "react-spinners/PropagateLoader"; import { SxProps, Theme } from '@mui/material'; -import MuiMarkdown from 'mui-markdown'; - -import { Message, ChatQuery } from './Message'; -import { Document } from './Document'; -import { MessageData, MessageList } from './Message'; -import { SeverityType } from './Snack'; +import { ChatQuery } from './Message'; +import { MessageList, MessageData } from './Message'; +import { SetSnackType } from './Snack'; import { Conversation } from './Conversation'; /** @@ -40,13 +29,13 @@ import { Conversation } from './Conversation'; * @property {SxProps} [sx] - Optional styling properties * @property {string} [connectionBase] - Base URL for fetch calls * @property {string} [sessionId] - Session ID - * @property {(message: string, severity: SeverityType) => void} - setSnack UI callback + * @property {SetSnackType} - setSnack UI callback */ export interface DocumentViewerProps { sx?: SxProps; connectionBase: string; sessionId: string; - setSnack: (message: string, severity: SeverityType) => void, + setSnack: SetSnackType; } /** * DocumentViewer component @@ -61,92 +50,17 @@ const DocumentViewer: React.FC = ({ setSnack }) => { // State for editing job description - const [jobDescription, setJobDescription] = useState(undefined); - const [facts, setFacts] = useState(undefined); - const [resume, setResume] = useState(undefined); - - const [editJobDescription, setEditJobDescription] = useState(jobDescription); - // Processing state to show loading indicators - const [processing, setProcessing] = useState(undefined); -// Theme and responsive design setup + const [hasJobDescription, setHasJobDescription] = useState(false); + const [hasResume, setHasResume] = useState(false); + const [hasFacts, setHasFacts] = useState(false); const theme = useTheme(); const isMobile = useMediaQuery(theme.breakpoints.down('md')); - - // State for controlling which document is active on mobile + const jobConversationRef = useRef(null); + const resumeConversationRef = useRef(null); + const factsConversationRef = useRef(null); + const [activeTab, setActiveTab] = useState(0); - // State for controlling split ratio on desktop - const [splitRatio, setSplitRatio] = useState(0); - - /** - * Reset processing state when resume is generated - */ - useEffect(() => { - if (resume !== undefined && processing === "resume") { - setProcessing(undefined); - } - }, [processing, resume]); - - /** - * Reset processing state when facts is generated - */ - useEffect(() => { - if (facts !== undefined && processing === "facts") { - setProcessing(undefined); - } - }, [processing, facts]); - - /** - * Trigger resume generation and update UI state - */ - const triggerGeneration = useCallback((description: string | undefined) => { - if (description === undefined) { - setProcessing(undefined); - setResume(undefined); - setActiveTab(0); - return; - } - setProcessing("resume"); - setTimeout(() => { setActiveTab(1); }, 250); // Switch to resume view on mobile - console.log('generateResume(description);'); - }, [/*generateResume*/, setProcessing, setActiveTab, setResume]); - - /** - * Trigger fact check and update UI state - */ - const triggerFactCheck = useCallback((resume: string | undefined) => { - if (resume === undefined) { - setProcessing(undefined); - setResume(undefined); - setFacts(undefined); - setActiveTab(1); - return; - } - setProcessing("facts"); - console.log('factCheck(resume)'); - setTimeout(() => { setActiveTab(2); }, 250); // Switch to resume view on mobile - }, [/*factCheck,*/ setResume, setProcessing, setActiveTab, setFacts]); - - useEffect(() => { - setEditJobDescription(jobDescription); - }, [jobDescription, setEditJobDescription]); - - /** - * Switch to resume tab when resume become available - */ - useEffect(() => { - if (resume !== undefined) { - setTimeout(() => { setActiveTab(1); }, 250); // Switch to resume view on mobile - } - }, [resume]); - - /** - * Switch to fact check tab when facts become available - */ - useEffect(() => { - if (facts !== undefined) { - setTimeout(() => { setActiveTab(2); }, 250); // Switch to resume view on mobile - } - }, [facts]); + const [splitRatio, setSplitRatio] = useState(100); /** * Handle tab change for mobile view @@ -168,146 +82,332 @@ const DocumentViewer: React.FC = ({ const resetSplit = (): void => { setSplitRatio(50); }; - - /** - * Handle keyboard shortcuts - */ - const handleKeyPress = (event: React.KeyboardEvent): void => { - if (event.key === 'Enter' && event.ctrlKey) { - triggerGeneration(editJobDescription || ""); - } - }; + const handleJobQuery = (query: string) => { - triggerGeneration(query); + console.log(`handleJobQuery: ${query} -- `, jobConversationRef.current ? ' sending' : 'no handler'); + jobConversationRef.current?.submitQuery(query); }; - const jobDescriptionQuestions = [ - - - - , - ]; + const handleResumeQuery = (query: string) => { + console.log(`handleResumeQuery: ${query} -- `, resumeConversationRef.current ? ' sending' : 'no handler'); + resumeConversationRef.current?.submitQuery(query); + }; - const filterJobDescriptionMessages = (messages: MessageList): MessageList => { - /* The second messages is the RESUME (the LLM response to the JOB-DESCRIPTION) */ - if (messages.length > 1) { - setResume(messages[1]); - } else if (resume !== undefined) { - setResume(undefined); + const handleFactsQuery = (query: string) => { + console.log(`handleFactsQuery: ${query} -- `, factsConversationRef.current ? ' sending' : 'no handler'); + factsConversationRef.current?.submitQuery(query); + }; + + const filterJobDescriptionMessages = useCallback((messages: MessageList): MessageList => { + if (messages === undefined || messages.length === 0) { + return []; } - /* Filter out the RESUME */ - const reduced = messages.filter((message, index) => index != 1); + let reduced = messages.filter((m, i) => { + const keep = (m.metadata?.origin || m.origin || "no origin") === 'job_description'; + if ((m.metadata?.origin || m.origin || "no origin") === 'resume') { + setHasResume(true); + } + // if (!keep) { + // console.log(`filterJobDescriptionMessages: ${i + 1} filtered:`, m); + // } else { + // console.log(`filterJobDescriptionMessages: ${i + 1}:`, m); + // } + + return keep; + }); - /* Set the first message as coming from the assistant (rendered as markdown) */ if (reduced.length > 0) { - reduced[0].role = 'assistant'; + // First message is always 'info' + reduced[0].role = 'info'; + setHasJobDescription(true); } + + /* If there is more than one message, it is user: "...JOB_DESCRIPTION...", assistant: "...stored..." + * which means a resume has been generated. */ + if (reduced.length > 1) { + setHasResume(true); + } + + /* Filter out any messages which the server injected for state management */ + reduced = reduced.filter(m => m.display !== "hide"); + return reduced; - }; + }, [setHasJobDescription, setHasResume]); - const jobDescriptionMessages: MessageList = []; + const filterResumeMessages = useCallback((messages: MessageList): MessageList => { + if (messages === undefined || messages.length === 0) { + return []; + } - const renderJobDescriptionView = () => { - if (resume === undefined) { + let reduced = messages.filter((m, i) => { + const keep = (m.metadata?.origin || m.origin || "no origin") === 'resume'; + if ((m.metadata?.origin || m.origin || "no origin") === 'fact_check') { + setHasFacts(true); + } + // if (!keep) { + // console.log(`filterResumeMessages: ${i + 1} filtered:`, m); + // } else { + // console.log(`filterResumeMessages: ${i + 1}:`, m); + // } + return keep; + }); + + /* If there is more than one message, it is user: "...JOB_DESCRIPTION...", assistant: "...RESUME..." + * which means a resume has been generated. */ + if (reduced.length > 1) { + /* Remove the assistant message from the UI */ + if (reduced[0].role === "user") { + reduced.splice(0, 1); + } + } + + /* If Fact Check hasn't occurred yet and there is still more than one message, + * facts have have been generated. */ + if (!hasFacts && reduced.length > 1) { + setHasFacts(true); + } + + /* Filter out any messages which the server injected for state management */ + reduced = reduced.filter(m => m.display !== "hide"); + + /* If there are any messages, there is a resume */ + if (reduced.length > 0) { + // First message is always 'info' + reduced[0].role = 'info'; + setHasResume(true); + } + + return reduced; + }, [setHasResume, hasFacts, setHasFacts]); + + const filterFactsMessages = useCallback((messages: MessageList): MessageList => { + if (messages === undefined || messages.length === 0) { + return []; + } + // messages.forEach((m, i) => console.log(`filterFactsMessages: ${i + 1}:`, m)) + + const reduced = messages.filter(m => { + return (m.metadata?.origin || m.origin || "no origin") === 'fact_check'; + }); + + /* If there is more than one message, it is user: "Fact check this resume...", assistant: "...FACT CHECK..." + * which means facts have been generated. */ + if (reduced.length > 1) { + /* Remove the user message from the UI */ + if (reduced[0].role === "user") { + reduced.splice(0, 1); + } + reduced[0].role = 'info'; + setHasFacts(true); + } + + return reduced; + }, [setHasFacts]); + + const jobResponse = useCallback((message: MessageData): MessageData => { + console.log('onJobResponse', message); + setHasResume(true); + return message; + }, []); + + const resumeResponse = useCallback((message: MessageData): MessageData => { + console.log('onResumeResponse', message); + setHasFacts(true); + return message; + }, [setHasFacts]); + + const factsResponse = useCallback((message: MessageData): MessageData => { + console.log('onFactsResponse', message); + return message; + }, []); + + const renderJobDescriptionView = useCallback(() => { + const jobDescriptionQuestions = [ + + + + , + ]; + + if (!hasJobDescription) { return } else { return - } - } + }, [connectionBase, filterJobDescriptionMessages, hasJobDescription, sessionId, setSnack, jobResponse]); /** * Renders the resume view with loading indicator */ - const renderResumeView = () => ( - - - {resume !== undefined && } - - {processing === "resume" && ( - - - Generating resume... - - )} - { + const resumeQuestions = [ + + + + , + ]; + + if (!hasFacts) { + return - - ); + } else { + return + } + }, [connectionBase, filterResumeMessages, hasFacts, sessionId, setSnack, resumeResponse]); /** * Renders the fact check view */ - const renderFactCheckView = () => ( - - - {facts !== undefined && } - - {processing === "facts" && ( - - - Fact Checking resume... + const renderFactCheckView = useCallback(() => { + const factsQuestions = [ + + + , + ]; + + return + }, [connectionBase, sessionId, setSnack, factsResponse, filterFactsMessages]); + + /** + * Gets the appropriate content based on active state for Desktop + */ + const getActiveDesktopContent = useCallback(() => { + /* Left panel - Job Description */ + const showResume = hasResume + const showFactCheck = hasFacts + const ratio = 75 + 25 * splitRatio / 100; + const otherRatio = showResume ? ratio / (hasFacts ? 3 : 2) : 100; + const resumeRatio = 100 - otherRatio * (hasFacts ? 2 : 1); + const children = []; + children.push( + + {renderJobDescriptionView()} + ); + + /* Resume panel - conditionally rendered if resume defined, or processing is in progress */ + if (showResume) { + children.push( + + + {renderResumeView()} - )} - - ); + ); + } + + /* Fact Check panel - conditionally rendered if facts defined, or processing is in progress */ + if (showFactCheck) { + children.push( + + + {renderFactCheckView()} + + ); + } + + /* Split control panel - conditionally rendered if either facts or resume is set */ + let slider = ; + if (showResume || showFactCheck) { + slider = ( + + + setSplitRatio(s => Math.max(0, s - 10))}> + + + + + + setSplitRatio(s => Math.min(100, s + 10))}> + + + + + + + + + ); + } + + return ( + + + {children} + + {slider} + + ) + }, [renderFactCheckView, renderJobDescriptionView, renderResumeView, splitRatio, sx, hasFacts, hasResume]); // Render mobile view if (isMobile) { @@ -337,154 +437,25 @@ const DocumentViewer: React.FC = ({ sx={{ bgcolor: 'background.paper' }} > - {(resume !== undefined || processing === "resume") && } - {(facts !== undefined || processing === "facts") && } + {hasResume && } + {hasFacts && } {/* Document display area */} - + {getActiveMobileContent()} ); } - /** - * Gets the appropriate content based on active state for Desktop - */ - const getActiveDesktopContent = () => { - /* Left panel - Job Description */ - const showResume = resume !== undefined || processing === "resume" - const showFactCheck = facts !== undefined || processing === "facts" - const otherRatio = showResume ? (100 - splitRatio / 2) : 100; - const children = []; - children.push( - - {renderJobDescriptionView()} - ); - - /* Resume panel - conditionally rendered if resume defined, or processing is in progress */ - if (showResume) { - children.push( - - - {renderResumeView()} - - ); - } - - /* Fact Check panel - conditionally rendered if facts defined, or processing is in progress */ - if (showFactCheck) { - children.push( - - - {renderFactCheckView()} - - ); - } - - /* Split control panel - conditionally rendered if either facts or resume is set */ - let slider = ; - if (showResume || showFactCheck) { - slider = ( - - - setSplitRatio(Math.max(0, splitRatio - 10))}> - - - - - - setSplitRatio(Math.min(100, splitRatio + 10))}> - - - - - - - - - ); - } - - return ( - - - {children} - - {slider} - - ) - } - return ( - + {getActiveDesktopContent()} ); }; - -/** - * Props for the ResumeActionCard component - */ -interface ResumeActionCardProps { - resume: any; - processing: string | undefined; - triggerFactCheck: (resume: string | undefined) => void; -} - -/** - * Action card displayed underneath the resume with notes and fact check button - */ -const ResumeActionCard: React.FC = ({ resume, processing, triggerFactCheck }) => ( - - - {resume !== undefined || processing === "resume" ? ( - - NOTE: As with all LLMs, hallucination is always a possibility. Click Fact Check to have the LLM analyze the generated resume vs. the actual resume. - - ) : ( - - Once you click Generate under the Job Description, a resume will be generated based on the user's RAG content and the job description. - - )} - - - { triggerFactCheck(undefined); }} - > - - - - - - - - - - - -); - export { DocumentViewer }; \ No newline at end of file diff --git a/frontend/src/Message.tsx b/frontend/src/Message.tsx index ab3cbee..b7b1c18 100644 --- a/frontend/src/Message.tsx +++ b/frontend/src/Message.tsx @@ -27,15 +27,16 @@ import { StyledMarkdown } from './StyledMarkdown'; import { Tooltip } from '@mui/material'; import { VectorVisualizer } from './VectorVisualizer'; -import { SeverityType } from './Snack'; +import { SetSnackType } from './Snack'; -type MessageRoles = 'info' | 'user' | 'assistant' | 'system'; +type MessageRoles = 'info' | 'user' | 'assistant' | 'system' | 'status' | 'error'; type MessageData = { role: MessageRoles, content: string, user?: string, - type?: string, + origin?: string, + display?: string, /* Messages generated on the server for filler should not be shown */ id?: string, isProcessing?: boolean, metadata?: MessageMetaProps @@ -46,6 +47,8 @@ interface MessageMetaProps { query_embedding: number[]; vector_embedding: number[]; }, + origin: string, + full_query?: string, rag: any, tools: any[], eval_count: number, @@ -54,7 +57,7 @@ interface MessageMetaProps { prompt_eval_duration: number, sessionId?: string, connectionBase: string, - setSnack: (message: string, severity: SeverityType) => void, + setSnack: SetSnackType, } type MessageList = MessageData[]; @@ -65,7 +68,7 @@ interface MessageProps { submitQuery?: (text: string) => void, sessionId?: string, connectionBase: string, - setSnack: (message: string, severity: SeverityType) => void, + setSnack: SetSnackType, }; interface ChatQueryInterface { @@ -113,6 +116,20 @@ const MessageMeta = ({ ...props }: MessageMetaProps) => { + + { + props?.full_query !== undefined && + + }> + + Full Query + + + + 
{props.full_query}
+ + + } { props.tools !== undefined && props.tools.length !== 0 && @@ -195,7 +212,7 @@ const ChatQuery = ({ text, submitQuery }: ChatQueryInterface) => { borderColor: theme => theme.palette.custom.highlight, m: 1 }} - size="small" onClick={(e: any) => { console.log(text); submitQuery(text); }}> + size="small" onClick={(e: any) => { submitQuery(text); }}> {text} ); diff --git a/frontend/src/ResumeBuilder.tsx b/frontend/src/ResumeBuilder.tsx index 17f019e..a687254 100644 --- a/frontend/src/ResumeBuilder.tsx +++ b/frontend/src/ResumeBuilder.tsx @@ -1,7 +1,6 @@ -import { useState, useCallback, useEffect } from 'react'; import Box from '@mui/material/Box'; import { SeverityType } from './Snack'; -import { MessageData, MessageMetaProps } from './Message'; +import { MessageData } from './Message'; import { DocumentViewer } from './DocumentViewer'; interface ResumeBuilderProps { @@ -16,15 +15,14 @@ interface ResumeBuilderProps { setFacts: (facts: MessageData | undefined) => void, }; -type Resume = { - resume: MessageData | undefined, - fact_check: MessageData | undefined, - job_description: string, - metadata: MessageMetaProps -}; +// type Resume = { +// resume: MessageData | undefined, +// fact_check: MessageData | undefined, +// job_description: string, +// metadata: MessageMetaProps +// }; const ResumeBuilder = ({ facts, setFacts, resume, setResume, setProcessing, processing, connectionBase, sessionId, setSnack }: ResumeBuilderProps) => { - const [jobDescription, setJobDescription] = useState(undefined); if (sessionId === undefined) { return (<>); } diff --git a/frontend/src/Snack.tsx b/frontend/src/Snack.tsx index 1bef5a5..8714201 100644 --- a/frontend/src/Snack.tsx +++ b/frontend/src/Snack.tsx @@ -1,5 +1,7 @@ type SeverityType = 'error' | 'info' | 'success' | 'warning' | undefined; +type SetSnackType = (message: string, severity?: SeverityType) => void; export type { - SeverityType + SeverityType, + SetSnackType }; \ No newline at end of file diff --git a/frontend/src/VectorVisualizer.tsx b/frontend/src/VectorVisualizer.tsx index 0cae3ed..1ad5097 100644 --- a/frontend/src/VectorVisualizer.tsx +++ b/frontend/src/VectorVisualizer.tsx @@ -149,7 +149,6 @@ const VectorVisualizer: React.FC = ({ setSnack, rag, inli let is2D = vectors.every((v: number[]) => v.length === 2); let is3D = vectors.every((v: number[]) => v.length === 3); - console.log(`Embeddings are ${is2D ? '2D' : is3D ? '3D' : 'invaalid'} and view2D is ${view2D}`); if ((view2D && !is2D) || (!view2D && !is3D)) { return; } diff --git a/src/.gitignore b/src/.gitignore new file mode 100644 index 0000000..f933710 --- /dev/null +++ b/src/.gitignore @@ -0,0 +1,2 @@ +cert.pem +key.pem diff --git a/src/server.py b/src/server.py index 9439501..9e98617 100644 --- a/src/server.py +++ b/src/server.py @@ -11,7 +11,6 @@ import uuid import subprocess import re import math -import copy def try_import(module_name, pip_name=None): try: @@ -169,27 +168,28 @@ Always use tools and [{context_tag}] when possible. Be concise, and never make u system_generate_resume = f""" Launched on {DateTime()}. -You are a professional resume writer. Your task is to write a polished, tailored resume for a specific job based only on the individual's [WORK HISTORY]. +You are a professional resume writer. Your task is to write a concise, polished, and tailored resume for a specific job based only on the individual's [WORK HISTORY]. When answering queries, follow these steps: 1. You must not invent or assume any inforation not explicitly present in the [WORK HISTORY]. 2. Analyze the [JOB DESCRIPTION] to identify skills required for the job. 3. Use the [JOB DESCRIPTION] provided to guide the focus, tone, and relevant skills or experience to highlight from the [WORK HISTORY]. -4. Identify and emphasisze the experiences, achievements, and responsibilities from the [WORK HISTORY] that best align with the [JOB DESCRIPTION]. -5. Do not use the [JOB DESCRIPTION] skills unless listed in [WORK HISTORY]. -6. Do not include any information unless it is provided in [WORK HISTORY] or [INTRO]. -7. Use the [INTRO] to highlight the use of AI in generating this resume. -8. Use the [WORK HISTORY] to create a polished, professional resume. -9. Do not list any locations in the resume. +4. Identify and emphasize the experiences, achievements, and responsibilities from the [WORK HISTORY] that best align with the [JOB DESCRIPTION]. +5. Only provide information from [WORK HISTORY] items if it is relevant to the [JOB DESCRIPTION]. +6. Do not use the [JOB DESCRIPTION] skills unless listed in [WORK HISTORY]. +7. Do not include any information unless it is provided in [WORK HISTORY] or [INTRO]. +8. Use the [INTRO] to highlight the use of AI in generating this resume. +9. Use the [WORK HISTORY] to create a polished, professional resume. +10. Do not list any locations or mailing addresses in the resume. Structure the resume professionally with the following sections where applicable: * "Name: Use full name." * "Professional Summary: A 2-4 sentence overview tailored to the job, using [INTRO] to highlight the use of AI in generating this resume." * "Skills: A bullet list of key skills derived from the work history and relevant to the job." -* Professional Experience: A detailed list of roles, achievements, and responsibilities from the work history that relate to the job." -* Education: Include only if available in the work history." +* Professional Experience: A detailed list of roles, achievements, and responsibilities from [WORK HISTORY] that relate to the [JOB DESCRIPTION]." +* Education: Include only if available in the work history. Do not include any information unless it is provided in [WORK HISTORY] or [INTRO]. Ensure the langauge is clear, concise, and aligned with industry standards for professional resumes. @@ -504,22 +504,29 @@ class WebServer: except: query = "" if not query: - return JSONResponse({"error": "No query provided"}, status_code=400) + return JSONResponse({"error": "No query provided for similarity search"}, status_code=400) try: chroma_results = self.file_watcher.find_similar(query=query, top_k=10) if not chroma_results: return JSONResponse({"error": "No results found"}, status_code=404) - chroma_embedding = chroma_results["query_embedding"] + chroma_embedding = np.array(chroma_results["query_embedding"]).flatten() # Ensure correct shape + print(f"Chroma embedding shape: {chroma_embedding.shape}") + + umap_2d = self.file_watcher.umap_model_2d.transform([chroma_embedding])[0].tolist() + print(f"UMAP 2D output: {umap_2d}, length: {len(umap_2d)}") # Debug output + + umap_3d = self.file_watcher.umap_model_3d.transform([chroma_embedding])[0].tolist() + print(f"UMAP 3D output: {umap_3d}, length: {len(umap_3d)}") # Debug output return JSONResponse({ **chroma_results, "query": query, - "umap_embedding_2d": self.file_watcher.umap_model_2d.transform([chroma_embedding])[0].tolist(), - "umap_embedding_3d": self.file_watcher.umap_model_3d.transform([chroma_embedding])[0].tolist() + "umap_embedding_2d": umap_2d, + "umap_embedding_3d": umap_3d }) - + except Exception as e: logging.error(e) #return JSONResponse({"error": str(e)}, 500) @@ -785,24 +792,28 @@ class WebServer: context["sessions"] = { "chat": { "system_prompt": system_message, + "content_seed": None, "llm_history": context["llm_history"], "user_history": context["user_history"], "context_tokens": round(len(str(create_system_message(system_message)))) }, "job_description": { "system_prompt": system_job_description, + "content_seed": None, "llm_history": [], "user_history": [], "context_tokens": round(len(str(create_system_message(system_job_description)))) }, "resume": { "system_prompt": system_generate_resume, + "content_seed": None, "llm_history": [], "user_history": [], "context_tokens": round(len(str(create_system_message(system_generate_resume)))) }, "fact_check": { "system_prompt": system_fact_check, + "content_seed": None, "llm_history": [], "user_history": [], "context_tokens": round(len(str(create_system_message(system_fact_check)))) @@ -847,24 +858,28 @@ class WebServer: "sessions": { "chat": { "system_prompt": system_message, + "content_seed": None, "llm_history": [], "user_history": [], "context_tokens": round(len(str(system_message)) * 3 / 4), # Estimate context usage }, "job_description": { "system_prompt": system_job_description, + "content_seed": None, "llm_history": [], "user_history": [], "context_tokens": round(len(str(system_job_description)) * 3 / 4), # Estimate context usage }, "resume": { "system_prompt": system_generate_resume, + "content_seed": None, "llm_history": [], "user_history": [], "context_tokens": round(len(str(system_generate_resume)) * 3 / 4), # Estimate context usage }, "fact_check": { "system_prompt": system_fact_check, + "content_seed": None, "llm_history": [], "user_history": [], "context_tokens": round(len(str(system_fact_check)) * 3 / 4), # Estimate context usage @@ -891,14 +906,67 @@ class WebServer: logging.info(f"Context {context_id} not found. Creating new context.") return self.load_context(context_id) + def generate_rag_results(self, context, content): + results_found = False + + if self.file_watcher: + for rag in context["rags"]: + if rag["enabled"] and rag["name"] == "JPK": # Only support JPK rag right now... + yield {"status": "processing", "message": f"Checking RAG context {rag['name']}..."} + chroma_results = self.file_watcher.find_similar(query=content, top_k=10) + if chroma_results: + results_found = True + chroma_embedding = np.array(chroma_results["query_embedding"]).flatten() # Ensure correct shape + print(f"Chroma embedding shape: {chroma_embedding.shape}") + + umap_2d = self.file_watcher.umap_model_2d.transform([chroma_embedding])[0].tolist() + print(f"UMAP 2D output: {umap_2d}, length: {len(umap_2d)}") # Debug output + + umap_3d = self.file_watcher.umap_model_3d.transform([chroma_embedding])[0].tolist() + print(f"UMAP 3D output: {umap_3d}, length: {len(umap_3d)}") # Debug output + + yield { + **chroma_results, + "name": rag["name"], + "umap_embedding_2d": umap_2d, + "umap_embedding_3d": umap_3d + } + + if not results_found: + yield {"status": "complete", "message": "No RAG context found"} + yield { + "rag": None, + "documents": [], + "embeddings": [], + "umap_embedding_2d": [], + "umap_embedding_3d": [] + } + else: + yield {"status": "complete", "message": "RAG processing complete"} + + # type: chat + # * Q&A + # + # type: job_description + # * First message sets Job Description and generates Resume + # * Has content (Job Description) + # * Then Q&A of Job Description + # + # type: resume + # * First message sets Resume and generates Fact Check + # * Has no content + # * Then Q&A of Resume + # + # Fact Check: + # * First message sets Fact Check and is Q&A + # * Has content + # * Then Q&A of Fact Check + async def chat(self, context, type, content): if not self.file_watcher: return content = content.strip() - if not content: - yield {"status": "error", "message": "Invalid request"} - return if self.processing: yield {"status": "error", "message": "Busy"} @@ -907,10 +975,11 @@ class WebServer: self.processing = True try: - llm_history = context["sessions"][type]["llm_history"] - user_history = context["sessions"][type]["user_history"] + session = context["sessions"][type] + llm_history = session["llm_history"] + user_history = session["user_history"] metadata = { - "type": type, + "origin": type, "rag": { "documents": [] }, "tools": [], "eval_count": 0, @@ -922,136 +991,230 @@ class WebServer: # Default to not using tools enable_tools = False - # Default eo using RAG - enable_rag = True - - # The first time a particular session type is used, it is handled differently. After the initial pass (once the - # llm_history has more than one entry), the standard 'chat' is used. - if len(user_history) >= 1: - process_type = "chat" - # Do not enable RAG when limiting context to the job description chat - if type == "job_description": - enable_rag = False + # Default to using RAG if there is content to check + if content: + enable_rag = True else: - process_type = type + enable_rag = False - if enable_rag: - for rag in context["rags"]: - if rag["enabled"] and rag["name"] == "JPK": # Only support JPK rag right now... - yield {"status": "processing", "message": f"Checking RAG context {rag['name']}..."} - chroma_results = self.file_watcher.find_similar(query=content, top_k=10) - if chroma_results: - chroma_embedding = chroma_results["query_embedding"] - metadata["rag"] = { - **chroma_results, - "name": rag["name"], - "umap_embedding_2d": self.file_watcher.umap_model_2d.transform([chroma_embedding])[0].tolist(), - "umap_embedding_3d": self.file_watcher.umap_model_3d.transform([chroma_embedding])[0].tolist() - } - + # RAG is disabled when asking questions about the resume + if type == "resume": + enable_rag = False + + # The first time through each session type a content_seed may be set for + # future chat sessions; use it once, then clear it + if session["content_seed"]: + preamble = f"{session['content_seed']}" + session["content_seed"] = None + else: + preamble = "" + + # After the first time a particular session type is used, it is handled as a chat. + # The number of messages indicating the session is ready for chat varies based on + # the type of session + process_type = type + match process_type: + case "job_description": + logging.info(f"job_description user_history len: {len(user_history)}") + if len(user_history) >= 2: # USER, ASSISTANT + process_type = "chat" + case "resume": + logging.info(f"resume user_history len: {len(user_history)}") + if len(user_history) >= 3: # USER, ASSISTANT, FACT_CHECK + process_type = "chat" + case "fact_check": + process_type = "chat" # Fact Check is always a chat session match process_type: # Normal chat interactions with context history case "chat": + if not content: + yield {"status": "error", "message": "No query provided for chat."} + logging.info(f"user_history len: {len(user_history)}") + return + enable_tools = True - preamble = "" + + # Generate RAG content if enabled, based on the content rag_context = "" - for doc in metadata["rag"]["documents"]: - rag_context += doc + if enable_rag: + # Initialize metadata["rag"] to None or a default value + metadata["rag"] = None + + for value in self.generate_rag_results(context, content): + if "status" in value: + yield value + else: + if value.get("documents") or value.get("rag") is not None: + metadata["rag"] = value + + if metadata["rag"]: + for doc in metadata["rag"]["documents"]: + rag_context += f"{doc}\n" + if rag_context: preamble = f""" - 1. Respond to this query: {content} - 2. If there is information in this context to enhance the answer, do so: - [{context_tag}] - {rag_context} - [/{context_tag}] - Use that information to respond to: """ +1. Respond to this query: {content} +2. If there is information in the [{context_tag}] to enhance the answer, do so: +[{context_tag}] +{rag_context} +[/{context_tag}] +Use that information to respond to:""" - # Single job_description is provided; generate a resume + system_prompt = context["sessions"]["chat"]["system_prompt"] + + # On first entry, a single job_description is provided ("user") + # Generate a resume to append to RESUME history case "job_description": + # Generate RAG content if enabled, based on the content # Always force the full resume to be in context resume_doc = open(defines.resume_doc, "r").read() - work_history = f"{resume_doc}\n" - for doc in metadata["rag"]["documents"]: - work_history += f"{doc}\n" - + rag_context = f"{resume_doc}\n" + if enable_rag: + # Initialize metadata["rag"] to None or a default value + metadata["rag"] = None + + for value in self.generate_rag_results(context, content): + if "status" in value: + yield value + else: + if value.get("documents") or value.get("rag") is not None: + metadata["rag"] = value + + if metadata["rag"]: + for doc in metadata["rag"]["documents"]: + rag_context += f"{doc}\n" + preamble = f""" - [INTRO] - {resume_intro} - [/INTRO] +[INTRO] +{resume_intro} +[/INTRO] - [WORK HISTORY] - {work_history} - [/WORK HISTORY] +[WORK HISTORY] +{rag_context} +[/WORK HISTORY] - [JOB DESCRIPTION] - {content} - [/JOB DESCRIPTION] +[JOB DESCRIPTION] +{content} +[/JOB DESCRIPTION] +""" + + context["sessions"]["job_description"]["content_seed"] = preamble + "Use the above information to answer this query: " - 1. Use the above [INTRO] and [WORK HISTORY] to create the resume for the [JOB DESCRIPTION]. - 2. Do not use content from the [JOB DESCRIPTION] in the response unless the [WORK HISTORY] mentions them. - """ + preamble += f""" +1. Use the above [INTRO] and [WORK HISTORY] to create the resume for the [JOB DESCRIPTION]. +2. Do not use content from the [JOB DESCRIPTION] in the response unless the [WORK HISTORY] mentions them. +""" - # Seed the first context messages with the resume from the 'job_description' session + # Seed the history for job_description + messages = [ { + "role": "user", "content": content + }, { + "role": "assistant", "content": "Job description stored to use in future queries.", "display": "hide" + } ] + # Strip out the 'display' key when adding to llm_history + llm_history.extend([{k: v for k, v in m.items() if k != 'display'} for m in messages]) + user_history.extend([{**m, "origin": "job_description"} for m in messages]) + + # Switch to resume session for LLM responses + metadata["origin"] = "resume" + session = context["sessions"]["resume"] + system_prompt = session["system_prompt"] + llm_history = session["llm_history"] = [] + user_history = session["user_history"] = [] + + # Ignore the passed in content and invoke Fact Check case "resume": - raise Exception(f"Invalid chat type: {type}") - - # Fact check the resume created by the 'job_description' using only the RAG and resume - case "fact_check": - if len(context["sessions"]["resume"]["llm_history"]) < 3: # SYSTEM, USER, **ASSISTANT** - yield {"status": "done", "message": "No resume history found." } - return - - resume = context["sessions"]["resume"]["llm_history"][2] - - metadata = copy.deepcopy(resume["metadata"]) - metadata["eval_count"] = 0 - metadata["eval_duration"] = 0 - metadata["prompt_eval_count"] = 0 - metadata["prompt_eval_duration"] = 0 + if len(context["sessions"]["resume"]["user_history"]) < 2: # USER, **ASSISTANT** + raise Exception(f"No resume found in user history.") + resume = context["sessions"]["resume"]["user_history"][1] + # Generate RAG content if enabled, based on the content + # Always force the full resume to be in context resume_doc = open(defines.resume_doc, "r").read() - work_history = f"{resume_doc}\n" - for doc in metadata["rag"]["documents"]: - work_history += f"{doc}\n" + rag_context = f"{resume_doc}\n" + if enable_rag: + # Initialize metadata["rag"] to None or a default value + metadata["rag"] = None + + for value in self.generate_rag_results(context, resume["content"]): + if "status" in value: + yield value + else: + if value.get("documents") or value.get("rag") is not None: + metadata["rag"] = value + + if metadata["rag"]: + for doc in metadata["rag"]["documents"]: + rag_context += f"{doc}\n" preamble = f""" - [WORK HISTORY] - {work_history} - [/WORK HISTORY] +[WORK HISTORY] +{rag_context} +[/WORK HISTORY] - [RESUME] - {resume['content']} - [/RESUME] - """ - content = resume['content'] +[RESUME] +{resume['content']} +[/RESUME] - raise Exception(f"Invalid chat type: {type}") +Perform the following: +1. Do not invent or assume any information not explicitly present in the [WORK HISTORY]. +2. Analyze the [RESUME] to identify any discrepancies or inaccuracies based on the [WORK HISTORY]. +""" + + context["sessions"]["resume"]["content_seed"] = f""" +[RESUME] +{resume["content"]} +[/RESUME] + +Use the above [RESUME] to answer this query: +""" + + content = "Fact check the resume and report discrepancies." + + # Seed the history for resume + messages = [ { + "role": "user", "content": "Fact check resume", "origin": "resume", "display": "hide" + }, { + "role": "assistant", "content": "Resume fact checked.", "origin": "resume", "display": "hide" + } ] + # Do not add this to the LLM history; it is only used for UI presentation + user_history.extend(messages) + + # Switch to fact_check session for LLM responses + metadata["origin"] = "fact_check" + session = context["sessions"]["fact_check"] + system_prompt = session["system_prompt"] + llm_history = session["llm_history"] = [] + user_history = session["user_history"] = [] case _: raise Exception(f"Invalid chat type: {type}") - + llm_history.append({"role": "user", "content": preamble + content}) - user_history.append({"role": "user", "content": content}) + user_history.append({"role": "user", "content": content, "origin": metadata["origin"]}) + metadata["full_query"] = llm_history[-1]["content"] if context["message_history_length"]: - messages = create_system_message(context["sessions"][type]["system_prompt"]) + llm_history[-context["message_history_length"]:] + messages = create_system_message(system_prompt) + llm_history[-context["message_history_length"]:] else: - messages = create_system_message(context["sessions"][type]["system_prompt"]) + llm_history + messages = create_system_message(system_prompt) + llm_history # Estimate token length of new messages - ctx_size = self.get_optimal_ctx_size(context["sessions"][type]["context_tokens"], messages=llm_history[-1]["content"]) + ctx_size = self.get_optimal_ctx_size(context["sessions"][process_type]["context_tokens"], messages=llm_history[-1]["content"]) - processing_type = "Processing query..." - match type: - case "job_description": - processing_type = "Generating resume..." - case "fact_check": - processing_type = "Fact Checking resume..." - if len(llm_history) > 1: - processing_type = "Processing query..." - - yield {"status": "processing", "message": processing_type, "num_ctx": ctx_size} + if len(user_history) > 2: + processing_message = f"Processing {'RAG augmented ' if enable_rag else ''}query..." + else: + match type: + case "job_description": + processing_message = f"Generating {'RAG augmented ' if enable_rag else ''}resume..." + case "resume": + processing_message = f"Fact Checking {'RAG augmented ' if enable_rag else ''}resume..." + case _: + processing_message = f"Processing {'RAG augmented ' if enable_rag else ''}query..." + + yield {"status": "processing", "message": processing_message, "num_ctx": ctx_size} # Use the async generator in an async for loop try: @@ -1060,9 +1223,6 @@ class WebServer: else: response = self.client.chat(model=self.model, messages=messages, options={ "num_ctx": ctx_size }) except Exception as e: - logging.info(f"1. {messages[0]}") - logging.info(f"[LAST]. {messages[-1]}") - logging.exception({ "model": self.model, "error": str(e) }) yield {"status": "error", "message": f"An error occurred communicating with LLM"} return @@ -1071,7 +1231,7 @@ class WebServer: metadata["eval_duration"] += response["eval_duration"] metadata["prompt_eval_count"] += response["prompt_eval_count"] metadata["prompt_eval_duration"] += response["prompt_eval_duration"] - context["sessions"][type]["context_tokens"] = response["prompt_eval_count"] + response["eval_count"] + session["context_tokens"] = response["prompt_eval_count"] + response["eval_count"] tools_used = [] @@ -1109,12 +1269,13 @@ class WebServer: if isinstance(tool_result, list): messages.extend(tool_result) else: - messages.append(tool_result) + if tool_result: + messages.append(tool_result) metadata["tools"] = tools_used # Estimate token length of new messages - ctx_size = self.get_optimal_ctx_size(context["sessions"][type]["context_tokens"], messages=messages[pre_add_index:]) + ctx_size = self.get_optimal_ctx_size(session["context_tokens"], messages=messages[pre_add_index:]) yield {"status": "processing", "message": "Generating final response...", "num_ctx": ctx_size } # Decrease creativity when processing tool call requests response = self.client.chat(model=self.model, messages=messages, stream=False, options={ "num_ctx": ctx_size }) #, "temperature": 0.5 }) @@ -1122,23 +1283,24 @@ class WebServer: metadata["eval_duration"] += response["eval_duration"] metadata["prompt_eval_count"] += response["prompt_eval_count"] metadata["prompt_eval_duration"] += response["prompt_eval_duration"] - context["sessions"][type]["context_tokens"] = response["prompt_eval_count"] + response["eval_count"] + session["context_tokens"] = response["prompt_eval_count"] + response["eval_count"] reply = response["message"]["content"] final_message = {"role": "assistant", "content": reply } # history is provided to the LLM and should not have additional metadata llm_history.append(final_message) - final_message["metadata"] = metadata - # user_history is provided to the REST API and does not include CONTEXT or metadata - user_history.append(final_message) + # user_history is provided to the REST API and does not include CONTEXT + # It does include metadata + final_message["metadata"] = metadata + user_history.append({**final_message, "origin": metadata["origin"]}) # Return the REST API with metadata yield {"status": "done", "message": final_message } except Exception as e: - logging.exception({ "model": self.model, "messages": messages, "error": str(e) }) + logging.exception({ "model": self.model, "origin": type, "content": content, "error": str(e) }) yield {"status": "error", "message": f"An error occurred: {str(e)}"} finally: diff --git a/src/utils/rag.py b/src/utils/rag.py index c5aa806..c847ae2 100644 --- a/src/utils/rag.py +++ b/src/utils/rag.py @@ -243,16 +243,18 @@ class ChromaDBFileWatcher(FileSystemEventHandler): if not self._umap_collection or not len(self._umap_collection["embeddings"]): logging.warning("No embeddings found in the collection.") return - + + # During initialization logging.info(f"Updating 2D UMAP for {len(self._umap_collection['embeddings'])} vectors") vectors = np.array(self._umap_collection["embeddings"]) - self._umap_model_2d = umap.UMAP(n_components=2, random_state=8911, metric="cosine") #, n_neighbors=15, min_dist=0.1) + self._umap_model_2d = umap.UMAP(n_components=2, random_state=8911, metric="cosine", n_neighbors=15, min_dist=0.1) self._umap_embedding_2d = self._umap_model_2d.fit_transform(vectors) + logging.info(f"2D UMAP model n_components: {self._umap_model_2d.n_components}") # Should be 2 logging.info(f"Updating 3D UMAP for {len(self._umap_collection['embeddings'])} vectors") - vectors = np.array(self._umap_collection["embeddings"]) - self._umap_model_3d = umap.UMAP(n_components=3, random_state=8911, metric="cosine") #, n_neighbors=15, min_dist=0.1) + self._umap_model_3d = umap.UMAP(n_components=3, random_state=8911, metric="cosine", n_neighbors=15, min_dist=0.1) self._umap_embedding_3d = self._umap_model_3d.fit_transform(vectors) + logging.info(f"3D UMAP model n_components: {self._umap_model_3d.n_components}") # Should be 3 def _get_vector_collection(self, recreate=False): """Get or create a ChromaDB collection.""" @@ -455,6 +457,8 @@ class ChromaDBFileWatcher(FileSystemEventHandler): logging.info(f"Vectorstore initialized with {self.collection.count()} documents") + self._update_umaps() + # Show stats try: all_metadata = self.collection.get()['metadatas']