[{"_1":2,"_177":-5,"_178":-5},"loaderData",{"_3":4,"_91":92},"root",{"_5":6,"_34":35,"_64":65},"ctx",{"_7":8,"_23":-5,"_24":25,"_26":27,"_28":20,"_29":20,"_30":20,"_31":32,"_33":20},"auth",{"_9":-5,"_10":-5,"_11":-5,"_12":-5,"_13":-5,"_14":-5,"_15":-5,"_16":-5,"_17":-5,"_18":-5,"_19":20,"_21":22},"sessionClaims","sessionId","sessionStatus","userId","actor","orgId","orgRole","orgSlug","orgPermissions","factorVerificationAge","isAuthenticated",false,"tokenType","session_token","user","preferencesKeybindings","default","preferencesTheme","light","preferencesFixers","preferencesCopilot","preferencesSelectionAi","notificationCount",0,"flagDenols","ENV",{"_36":37,"_38":39,"_40":41,"_42":43,"_44":45,"_46":47,"_48":49,"_50":51,"_52":53,"_54":55,"_56":57,"_58":59,"_60":61,"_62":63},"REACT_APP_SENTRY_DSN","https://586020d86f05489db63a554690acf1e9@o1398359.ingest.sentry.io/6724499","VAL_RUNTIME_NAME","prod","RENDER_GIT_COMMIT","4ff166458996ccfe3bfa8627489090938620785a","RENDER_GIT_BRANCH","main","CLERK_PUBLISHABLE_KEY","pk_live_Y2xlcmsudmFsLnRvd24k","POSTHOG_API_TOKEN","phc_nmoyuAkStEZuUp69rYlg1DZgrIt19Gx0Y4Hwf8tGgJP","POSTHOG_API_HOST","https://wart.val.town","WEB_API_TEMPLATE","https://{val}.web.val.run","APP_URL","https://www.val.town","MODULE_URL","https://esm.town","VANITY_WEB_API_TEMPLATE","https://{label}.val.run/","DENOLS_URL","https://lsp.val.town","NODE_ENV","production","API_URL","https://api.val.town","clerkState",{"_66":67},"__internal_clerk_state",{"_68":69,"_70":45,"_71":72,"_73":72,"_74":20,"_75":72,"_76":72,"_77":72,"_78":72,"_79":80,"_89":20,"_90":20},"__clerk_ssr_state",{"_21":22,"_9":-5,"_10":-5,"_11":-5,"_12":-5,"_13":-5,"_14":-5,"_15":-5,"_16":-5,"_17":-5,"_18":-5,"_19":20},"__publishableKey","__proxyUrl","","__domain","__isSatellite","__signInUrl","__signUpUrl","__afterSignInUrl","__afterSignUpUrl","__clerk_debug",{"_81":20,"_19":20,"_82":72,"_83":84,"_85":72,"_86":45,"_87":20,"_88":72},"isSignedIn","proxyUrl","reason","session-token-and-uat-missing","message","publishableKey","isSatellite","domain","__telemetryDisabled","__telemetryDebug","routes/embed.x.$handle.$name.$",{"_93":94,"_95":96,"_97":98,"_99":100,"_150":151,"_174":175,"_176":-5},"handle","lightweight","name","findings","path","backend/services/openai.service.ts","project",{"_101":102,"_95":96,"_103":104,"_105":-5,"_106":20,"_107":108,"_109":-5,"_110":111,"_12":112,"_113":-5,"_114":20,"_115":-5,"_116":32,"_14":117,"_118":119,"_120":119,"_121":122,"_123":122,"_124":122,"_125":119,"_126":119,"_23":127,"_130":131,"_133":134,"_149":20},"id","d0eeb5a0-7951-11f0-8810-0224a6c84d84","createdAt",["D",1755205275298],"deletedAt","legacyEnvVars","privacy","public","description","imageUrl","https://imagedelivery.net/iHX6Ovru0O7AjmyT5yZRoA/c62a156b-37c5-46ed-47f2-b6fdd4bd1a00/square512","449588c0-326d-11f0-a80c-0741f032ef73","pinnedFileId","pinPreview","screenshotVersion","likeCount","65f1d616-c14b-419c-88a0-e2308cab925b","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm",{"_93":94,"_128":129},"profileImageUrl","https://imagedelivery.net/iHX6Ovru0O7AjmyT5yZRoA/aa289d5e-22db-464f-ba88-7288e2183b00/square512","org",{"_93":94,"_132":129},"avatar","branch",{"_135":136,"_95":43,"_101":137,"_138":102,"_139":140,"_141":142,"_143":108,"_144":145,"_146":117,"_147":96,"_148":94},"version",46,"d0f324fa-7951-11f0-8810-0224a6c84d84","projectId","forkedBranchId","2af6b7a2-790f-11f0-b3db-0224a6c84d84","forkedBranchVersion",173,"forkedProjectPrivacy","forkedProjectId","2af64cf4-790f-11f0-b3db-0224a6c84d84","forkedProjectOrgId","projectName","userHandle","isOwned","file",{"_101":152,"_95":153,"_154":155,"_156":157,"_158":137,"_159":160,"_161":162,"_103":163,"_135":164,"_165":166,"_167":168,"_169":20,"_170":-5,"_97":171},"48eaad56-7a0c-11f0-b82f-0224a6c84d84","openai.service.ts","fileId","82c4626c-7a05-11f0-9236-0224a6c84d84","parentId","0198aa63-0ee2-725e-8c9f-e39dc7a1bd77","branchId","branchVersion",23,"latest",true,["D",1755285362994],1,"fileType","script","content","/**\n * Two-Stage Consensus Analysis for Meeting Transcripts\n * \n * Stage 1: Analyze each transcript individually against the question\n * Stage 2: Synthesize consensus across all individual analyses\n * \n * Uses actual OpenAI Chat Completions API with proper error handling\n */\n\nimport { OpenAI } from \"https://esm.town/v/std/openai\";\n\n/*** Types ***/\nexport interface ConsensusAnalysisRequest {\n question: string;\n transcripts: string[];\n apiKey: string;\n}\n\nexport type Stance = \"supports\" | \"mixed\" | \"contradicts\" | \"not enough evidence\";\n\nexport interface PerTranscriptFinding {\n key: string;\n stance: Stance;\n shortRationale: string; // <= 120 chars\n}\n\nexport interface ConsensusAnswer {\n question: string;\n consensusLevel: \"high\" | \"medium\" | \"low\";\n directAnswer: string;\n keyPoints: string[];\n disagreements: string[];\n confidence: number; // 0..1\n methodologyNote: string;\n perTranscriptFindings: PerTranscriptFinding[];\n}\n\nexport interface ConsensusAnalysisResponse {\n success: boolean;\n data?: ConsensusAnswer;\n error?: string;\n timestamp: string;\n}\n\nexport interface ConsensusOptions {\n extractModel?: string; // default: \"gpt-4o-mini\"\n consensusModel?: string; // default: \"gpt-4o\"\n maxConcurrency?: number; // default: 3\n maxTranscriptChars?: number; // default: 15000\n}\n\ninterface IndividualAnalysis {\n transcriptKey: string;\n stance: Stance;\n keyFindings: string[];\n supportingQuotes: string[];\n rationale: string;\n confidence: number;\n}\n\n/*** Configuration ***/\nconst DEFAULTS: Required = {\n extractModel: \"gpt-4o-mini\",\n consensusModel: \"gpt-4o\",\n maxConcurrency: 3,\n maxTranscriptChars: 100000,\n};\n\n/*** Stage 1: Individual Transcript Analysis ***/\nconst INDIVIDUAL_ANALYSIS_PROMPT = `You are a meticulous research analyst. Analyze the provided interview transcript to answer a specific research question.\n\nYour task:\n1. Determine your stance on how this transcript answers the question\n2. Extract key findings that support your stance\n3. Find relevant quotes (if any) that support your findings\n4. Provide a clear rationale for your stance\n5. Rate your confidence in this analysis\n\nResponse format (JSON):\n{\n \"stance\": \"supports|mixed|contradicts|not enough evidence\",\n \"keyFindings\": [\"finding 1\", \"finding 2\", \"finding 3\"],\n \"supportingQuotes\": [\"quote 1\", \"quote 2\"],\n \"rationale\": \"Brief explanation of why you chose this stance\",\n \"confidence\": 0.85\n}\n\nGuidelines:\n- Be objective and evidence-based\n- If the transcript doesn't address the question, use \"not enough evidence\"\n- Keep findings concise but specific\n- Quotes should be under 140 characters\n- Confidence should reflect how clearly the transcript addresses the question (0.0 to 1.0)`;\n\nasync function analyzeIndividualTranscript(\n openai: OpenAI,\n model: string,\n question: string,\n transcript: string,\n transcriptKey: string,\n maxChars: number\n): Promise {\n // Truncate transcript if too long\n const truncatedTranscript = transcript.length > maxChars \n ? transcript.substring(0, maxChars) + \"... [truncated]\"\n : transcript;\n\n const userPrompt = `RESEARCH QUESTION:\n${question}\n\nTRANSCRIPT (${transcriptKey}):\n${truncatedTranscript}`;\n\n try {\n const completion = await openai.chat.completions.create({\n model,\n messages: [\n { role: \"system\", content: INDIVIDUAL_ANALYSIS_PROMPT },\n { role: \"user\", content: userPrompt }\n ],\n response_format: { type: \"json_object\" },\n temperature: 0.1,\n max_tokens: 1000\n });\n\n const content = completion.choices[0]?.message?.content;\n if (!content) {\n throw new Error(\"No response content from OpenAI\");\n }\n\n const parsed = JSON.parse(content);\n \n return {\n transcriptKey,\n stance: parsed.stance || \"not enough evidence\",\n keyFindings: Array.isArray(parsed.keyFindings) ? parsed.keyFindings : [],\n supportingQuotes: Array.isArray(parsed.supportingQuotes) ? parsed.supportingQuotes : [],\n rationale: parsed.rationale || \"No rationale provided\",\n confidence: typeof parsed.confidence === 'number' ? parsed.confidence : 0.5\n };\n\n } catch (error) {\n console.error(`Error analyzing ${transcriptKey}:`, error);\n return {\n transcriptKey,\n stance: \"not enough evidence\",\n keyFindings: [],\n supportingQuotes: [],\n rationale: `Analysis failed: ${error.message}`,\n confidence: 0.0\n };\n }\n}\n\n/*** Stage 2: Consensus Synthesis ***/\nconst CONSENSUS_SYNTHESIS_PROMPT = `You are a senior research synthesizer. You will receive multiple individual analyses of different transcripts for the same research question. Your job is to synthesize these into a comprehensive consensus answer.\n\nAnalyze the individual findings and create a consensus that:\n1. Provides a direct, actionable answer to the research question\n2. Identifies the level of consensus across transcripts\n3. Highlights key points that most transcripts agree on\n4. Notes any significant disagreements or outliers\n5. Provides an overall confidence score\n6. Explains your methodology\n\nResponse format (JSON):\n{\n \"consensusLevel\": \"high|medium|low\",\n \"directAnswer\": \"Clear, direct answer to the research question\",\n \"keyPoints\": [\"point 1\", \"point 2\", \"point 3\"],\n \"disagreements\": [\"disagreement 1\", \"disagreement 2\"],\n \"confidence\": 0.85,\n \"methodologyNote\": \"Brief explanation of how you reached consensus\"\n}\n\nConsensus Level Guidelines:\n- HIGH: 80%+ of transcripts support the same conclusion\n- MEDIUM: 60-79% agreement, or mixed but trending\n- LOW: <60% agreement, or significant contradictions\n\nConfidence Guidelines:\n- Consider both the consensus level and the quality of evidence\n- Factor in how directly transcripts addressed the question\n- Account for the strength of supporting evidence`;\n\nasync function synthesizeConsensus(\n openai: OpenAI,\n model: string,\n question: string,\n individualAnalyses: IndividualAnalysis[]\n): Promise {\n const analysisText = individualAnalyses.map((analysis, index) => \n `### ${analysis.transcriptKey}\nStance: ${analysis.stance}\nConfidence: ${analysis.confidence}\nKey Findings: ${analysis.keyFindings.join('; ')}\nSupporting Quotes: ${analysis.supportingQuotes.join('; ')}\nRationale: ${analysis.rationale}`\n ).join('\\n\\n');\n\n const userPrompt = `RESEARCH QUESTION:\n${question}\n\nINDIVIDUAL TRANSCRIPT ANALYSES:\n${analysisText}\n\nPlease synthesize these analyses into a comprehensive consensus answer.`;\n\n try {\n const completion = await openai.chat.completions.create({\n model,\n messages: [\n { role: \"system\", content: CONSENSUS_SYNTHESIS_PROMPT },\n { role: \"user\", content: userPrompt }\n ],\n response_format: { type: \"json_object\" },\n temperature: 0.1,\n max_tokens: 1500\n });\n\n const content = completion.choices[0]?.message?.content;\n if (!content) {\n throw new Error(\"No response content from OpenAI\");\n }\n\n const parsed = JSON.parse(content);\n \n // Create per-transcript findings for the final output\n const perTranscriptFindings: PerTranscriptFinding[] = individualAnalyses.map(analysis => ({\n key: analysis.transcriptKey,\n stance: analysis.stance,\n shortRationale: analysis.rationale.length > 120 \n ? analysis.rationale.substring(0, 117) + \"...\"\n : analysis.rationale\n }));\n\n return {\n question,\n consensusLevel: parsed.consensusLevel || \"low\",\n directAnswer: parsed.directAnswer || \"Unable to determine consensus answer\",\n keyPoints: Array.isArray(parsed.keyPoints) ? parsed.keyPoints : [],\n disagreements: Array.isArray(parsed.disagreements) ? parsed.disagreements : [],\n confidence: typeof parsed.confidence === 'number' ? parsed.confidence : 0.5,\n methodologyNote: parsed.methodologyNote || \"Consensus analysis completed\",\n perTranscriptFindings\n };\n\n } catch (error) {\n console.error(\"Error in consensus synthesis:\", error);\n \n // Fallback consensus based on individual analyses\n const stanceCounts = individualAnalyses.reduce((acc, analysis) => {\n acc[analysis.stance] = (acc[analysis.stance] || 0) + 1;\n return acc;\n }, {} as Record);\n\n const totalAnalyses = individualAnalyses.length;\n const maxStance = Object.entries(stanceCounts).reduce((a, b) => \n stanceCounts[a[0] as Stance] > stanceCounts[b[0] as Stance] ? a : b\n )[0] as Stance;\n \n const maxCount = stanceCounts[maxStance];\n const consensusPercentage = maxCount / totalAnalyses;\n \n const consensusLevel = consensusPercentage >= 0.8 ? \"high\" : \n consensusPercentage >= 0.6 ? \"medium\" : \"low\";\n\n const perTranscriptFindings: PerTranscriptFinding[] = individualAnalyses.map(analysis => ({\n key: analysis.transcriptKey,\n stance: analysis.stance,\n shortRationale: analysis.rationale.length > 120 \n ? analysis.rationale.substring(0, 117) + \"...\"\n : analysis.rationale\n }));\n\n return {\n question,\n consensusLevel,\n directAnswer: `Synthesis failed, but ${maxCount}/${totalAnalyses} transcripts ${maxStance} the question. Error: ${error.message}`,\n keyPoints: [],\n disagreements: [`Synthesis error: ${error.message}`],\n confidence: 0.3,\n methodologyNote: `Fallback analysis due to synthesis error. Based on stance distribution: ${JSON.stringify(stanceCounts)}`,\n perTranscriptFindings\n };\n }\n}\n\n/*** Concurrency Control ***/\nasync function mapWithLimit(\n items: T[],\n limit: number,\n fn: (item: T, index: number) => Promise\n): Promise {\n const results: R[] = new Array(items.length);\n let index = 0;\n\n const workers = Array.from({ length: Math.min(limit, items.length) }, async () => {\n while (index < items.length) {\n const currentIndex = index++;\n results[currentIndex] = await fn(items[currentIndex], currentIndex);\n }\n });\n\n await Promise.all(workers);\n return results;\n}\n\n/*** Public API ***/\nexport async function analyzeTranscriptsWithConsensus(\n request: ConsensusAnalysisRequest,\n options: ConsensusOptions = {}\n): Promise {\n const timestamp = new Date().toISOString();\n\n // Validation\n if (!request?.apiKey) {\n return { success: false, error: \"Missing OpenAI API key\", timestamp };\n }\n if (!request?.question?.trim()) {\n return { success: false, error: \"Missing or empty question\", timestamp };\n }\n if (!request?.transcripts?.length) {\n return { success: false, error: \"No transcripts provided\", timestamp };\n }\n\n const config = { ...DEFAULTS, ...options };\n const openai = new OpenAI({ apiKey: request.apiKey });\n\n try {\n console.log(`🔍 Starting 2-stage analysis for question: \"${request.question}\"`);\n console.log(`📄 Processing ${request.transcripts.length} transcripts with max ${config.maxConcurrency} concurrent requests`);\n\n // Stage 1: Analyze each transcript individually\n console.log(\"📊 Stage 1: Individual transcript analysis...\");\n const individualAnalyses = await mapWithLimit(\n request.transcripts,\n config.maxConcurrency,\n async (transcript, index) => {\n const transcriptKey = `Transcript ${index + 1}`;\n console.log(` 🔍 Analyzing ${transcriptKey}...`);\n \n const analysis = await analyzeIndividualTranscript(\n openai,\n config.extractModel,\n request.question,\n transcript,\n transcriptKey,\n config.maxTranscriptChars\n );\n \n console.log(` ✅ ${transcriptKey}: ${analysis.stance} (confidence: ${Math.round(analysis.confidence * 100)}%)`);\n return analysis;\n }\n );\n\n // Stage 2: Synthesize consensus\n console.log(\"🤝 Stage 2: Consensus synthesis...\");\n const consensus = await synthesizeConsensus(\n openai,\n config.consensusModel,\n request.question,\n individualAnalyses\n );\n\n console.log(`✅ Consensus reached: ${consensus.consensusLevel} (confidence: ${Math.round(consensus.confidence * 100)}%)`);\n\n return {\n success: true,\n data: consensus,\n timestamp\n };\n\n } catch (error) {\n console.error(\"❌ Error in consensus analysis:\", error);\n return {\n success: false,\n error: error instanceof Error ? error.message : String(error),\n timestamp\n };\n }\n}\n\n// Legacy function for backward compatibility (not used in new approach)\nexport async function analyzeTranscriptsForQuestion(\n question: string,\n transcripts: string[],\n apiKey: string\n): Promise<{ success: boolean; data?: any; error?: string }> {\n const result = await analyzeTranscriptsWithConsensus({\n question,\n transcripts,\n apiKey\n });\n \n return {\n success: result.success,\n data: result.data ? { answer: result.data.directAnswer } : undefined,\n error: result.error\n };\n}","deleted","label",[172,173,153],"backend","services","permalink","https://www.val.town/x/lightweight/findings/code/backend/services/openai.service.ts","extension","actionData","errors"]