AGENTIC RESEARCH SYSTEM
Methodic Replay
A multi-agent research pipeline with calibrated confidence, gap detection, and autonomous re-planning. +0.692 composite quality vs static baseline.
PROBLEM
Static survey forms collect research data field-by-field, with no way to notice mid-study that a question area is going unanswered. In the recorded benchmark, the static baseline left most of the eight target variables uncovered (Methodic's coverage delta: +0.833).
APPROACH
Methodic uses a multi-agent pipeline: a clarification agent scopes the study, conversation agents conduct structured sessions, a coverage checker detects gaps after each session, and an autonomous re-planner adds reserve participants when a variable remains unresolved.
MY ROLE
Solo builder (product + engineering). Defined the research variables, conversation protocol, coverage rubric, and scoring dimensions. Implemented all agents, the deterministic fixture replay, and the self-benchmarking quality layer.
OUTCOME
+0.692 composite quality uplift vs static baseline (0.761 vs 0.069). Coverage 100% vs 16.7% for the static form. Built and deployed to Cloud Run for the Google AI Agent Challenge.
HONEST CAVEATS
The comparison is self-benchmarked — the static baseline is a reference fixture, not a measured production study. Participants are simulated Gemini personas; results reflect fixture-driven deterministic replay, not live NLP extraction.
REPLAY DEMO
Agent Timeline
Brief received
A sales-insights agent submits a research request: ‘Should we adjust mid-market enterprise packaging or ROI messaging?’ Methodic clarifies study scope before proceeding.
COVERAGE HEATMAP — 0 / 4 participants
No sessions run yet.
Deterministic replay of a recorded run. Participants are simulated personas (Gemini fixture data). Scoring is self-benchmarked against a static-form baseline.