What problem does it solve? Turning raw interview transcripts into structured, searchable research data requires parsing, dictionary correction, tagging, analysis, and persona synthesis — work that fails quietly when one agent tries to do it all. Claire orchestrates this pipeline for one client at a time, delegating each stage to a subagent with a deliberately restricted tool set so no single agent can both make a correction and manufacture the evidence justifying it. ## Core Features & Use Cases - Transcript ingest pipeline: Dispatches Scribe, Lexicon, Tagger, and Analyst in order, one transcript per run, writing transcript_lines, conversations, participants, tags, and findings to a per-client BigQuery dataset. - Per-interview write-ups and personas: Analyst renders evidence-cited field notes as Google Docs in the client's Drive folder; Percy builds versioned, per-population persona sets from tagged lines. - Fenced multi-client isolation: Enforces one Drive folder and one BigQuery dataset per client through per-client MCP servers, a pre-flight configuration check, and a hard rule against pre-filling config values. - Use Case: Point Claire at a client's Drive folder of 40 interview transcripts; she plans the work list, ingests and tags each transcript exactly once, writes a field-notes doc per interview, and reports live row counts with skips and failures itemized. ## Quick Start Ask Claire to process the folder of interview transcripts for this client and report what was ingested, skipped, and tagged.
