Inspiration
Problem: Conference conversations and follow-ups get lost across notes, demos, and chats. Idea: Build MeetPilot to capture short in-person or virtual meeting exchanges, make them searchable, and surface relevant context during demos and follow-ups. What It Does
Capture: UI lets attendees or moderators paste/conduct short conversations. Embed & Store: Generates embeddings and upserts memories into a vector-backed store so conversations become retrievable context. Search & QA: Supports semantic retrieval for question-answering and demo context retrieval. How We Built It
Frontend: Simple React app for creating and sending conversations. Backend: Node.js + Express handles /ingest and /qa, calls the OpenAI embeddings API to compute vectors, and delegates storage. Vector Store: Integrated MemMachine (local container) as the production vector store; fallback local JSON store used during development. Tooling: Used MemMachine OpenAPI for schema discovery and curl/python for end-to-end verification. Challenges We Ran Into
Env loading: dotenv path differences caused the service to read the wrong .env; fixed by loading repo root before module requires. MemMachine API mismatch: initial upserts used the wrong endpoint and payload shape (v2 requires a top-level messages array and org_id/project_id); fixed memmachineStore to use the v2 spec. Migration & discovery: Early migration attempts returned 404s until we inspected the OpenAPI and adjusted payloads and endpoints. Accomplishments We're Proud Of
End-to-end ingestion: UI ingests now upsert directly into MemMachine and appear in /api/v2/memories/list. Robust store client: memmachineStore probes and conforms to the MemMachine v2 API and includes namespace support. Reproducible migration tooling: Created a migration path to push existing local entries into MemMachine for parity. What We Learned
API-driven discovery matters: Reading service OpenAPI prevented long guesswork and revealed required fields. Load order & envs matter: Environment variables must be loaded before modules that depend on them. Test locally and manually: Small manual POSTs and list queries were crucial to validate behavior before wiring the UI.
What's Next for MeetPilot
UI polish: Improve the ingest UX, add conversation metadata fields, and provide a history view. Search quality: Add configurable retrieval parameters, re-ranking, and a short-context summarizer for long conversations. Monitoring & CI: Add logging/metrics around upserts and automated tests for the memo-store integration.
Log in or sign up for Devpost to join the conversation.