Feat: Enhance query intent search and answer format readability

2026-05-02 18:20:22 +09:00
parent 26bbb22c7e
commit da4ebe3942
9 changed files with 173 additions and 12 deletions
@@ -0,0 +1,35 @@
+# Development Log: Query Intent Search Tuning
+
+## Purpose
+Improve Second Brain search relevance and answer framing for questions about customer evaluation, approval likelihood, UX, business value, and requirement fit.
+
+## User Feedback
+The answer became safer, but Second Brain still retrieved architecture documents for a question that was really about customer experience, approval risk, and business direction. Approval likelihood should also be labeled as inference, not fact.
+
+## Implementation Summary
+- Added `queryIntent` classification to Second Brain Trace:
+  - `technical`
+  - `ux-business`
+  - `governance`
+  - `general`
+- Expanded retrieval query terms based on intent.
+- For `ux-business`, boosted UX, customer journey, product discovery, virtual store, stakeholder approval, requirement fit, business value, acceptance criteria, and conversion-flow documents.
+- For `ux-business`, penalized API Gateway, microservice, monolithic, backend, database, routing, and generic architecture-principles paths.
+- Added Trace context guidance that approval likelihood is inference unless explicit approval criteria are provided.
+- Added base and Guard prompt guidance to prefer UX/business framing over technical architecture for these questions.
+- Added tests to verify UX/business docs outrank API Gateway docs for approval/customer-experience questions.
+
+## Changed Files
+- `src/features/secondBrainTrace.ts`
+- `src/features/projectChronicle/guardPrompt.ts`
+- `src/utils.ts`
+- `tests/secondBrainTrace.test.ts`
+- `tests/projectChronicleGuardPrompt.test.ts`
+
+## Verification
+- `./node_modules/.bin/tsc --noEmit`
+- `npm run compile`
+- `./node_modules/.bin/jest --runInBand`
+
+## Result
+Second Brain Trace should now better match the user's question intent. Customer approval and UX/business-fit questions should retrieve customer journey and business-value notes before generic technical architecture notes.