What it does
User Interview Synthesizer takes the three to four hours of manual analysis and tagging that follow every one-hour interview and turns them into a report you can hand to a stakeholder. You supply coded interview transcripts — segments with a timestamp, the text, a theme code and a stance — plus the research questions you were trying to answer. The agent aggregates theme codes with per-participant frequency, then builds an evidence table organised by research question, with every quote timestamped and attributed so a reader can trace any claim back to who said it and when. The confirmation-bias guardrail is the reason to choose this over a summariser. Any research question that has both supporting and contradicting evidence is flagged and listed in a separate counter-evidence section — never folded into the narrative where a dissenting voice quietly disappears. `list_counter_evidence` returns that section on its own, which is a useful thing to read first rather than last. Per-participant frequency matters for the same reason. A theme mentioned nine times by one participant is not the same finding as one mentioned once by nine, and the report keeps those distinguishable. Each step is callable alone. `aggregate_themes` gives the theme counts. `build_evidence_table` gives the quotes per question. `synthesize_report` runs the entire deterministic synthesis and returns it without any language-model narration at all — which is what you want when the output is going into a document you will write yourself. Everything numeric and every quote comes from the corpus you provide; the model only narrates. Quotes are sanitised for prompt injection before they reach it, and suspicious segments are flagged. No connectors and no credentials. Transcripts from Zoom or Gong need exporting and coding yourself before you pass them in — the agent does not fetch or transcribe. Built for UX researchers and product managers.
Example prompts
- Synthesize these 12 coded interviews into a research report
- Which themes came up most, and across how many participants?
- Show me the evidence for each research question
