001 / Builds
← Back to BuildsUX Synthesizer
AI-assisted qualitative UX research synthesis: ingest transcripts, notes, and survey verbatims and produce structured insights - themes, pain points, personas, codebook, and recommendations - with grounding checks and a transcript cleanup pipeline.
Stack
Next.js · TypeScript · OpenAI · Zod · Tailwind
Key Decision
Zod-validated analysis JSON plus deterministic transcript normalization, with optional LLM repair and evidence repair for ungrounded snippets.
Outcome
Researchers move from raw qualitative data to schema-valid, quote-backed outputs they can review, copy, and export without losing traceability to the source.
Proof System
Vitest and Playwright coverage in CI, grounding checks for every generated claim, and recovery paths when transcript quality is low.
System
- Ingests transcripts, notes, and survey verbatims into a normalized analysis pipeline.
- Produces structured synthesis outputs: themes, pain points, personas, codebook, and recommendations.
- Runs grounding checks on generated claims and repairs ungrounded evidence snippets.
- Supports deterministic transcript cleanup with optional LLM-assisted repair when input quality is low.