Maitch

MAITCH is a B2B/B2B2C matchmaking platform for event organizers. Upload a guest list (CSV), we enrich profiles with public web signals, vector-match them in Qdrant, and surface high-quality “who should meet whom” suggestions with clear rationales. The goal: make in-person networking efficient, measurable, and delightful for hackathons, conferences, and corporate events.

How we meet the judging criteria

Execution & Functionality
Working MVP: CSV upload → light enrichment → vectorization → Qdrant similarity search → ranked suggestions with justifications. Deployed on Vercel with a clean Next.js UI and Firebase for persistence/auth.

AI & Agents Usage
Today: deterministic pipeline + embeddings + vector search (no autonomous agents). Architecture is agent-ready: an orchestration layer can assign tasks (enrich → dedupe → embed → match → notify) with retries, rate-limits, and human-in-the-loop review.

Scalability (One-Person Model)
A single organizer can process 1,000+ attendees: batch ingestion, automatic enrichment, and hands-off matching. Qdrant scales horizontally; stateless Next.js API routes and Firebase keep ops minimal.

Problem Clarity & Market Impact
IRL networking is noisy and inefficient; organizers lack tools that translate attendee data into targeted introductions. MAITCH improves attendee satisfaction, partner value, and measurable outcomes (intros made, meetings booked).

Demo & Product Narrative
Organizer uploads CSV.
System enriches and normalizes profiles.
We embed and index in Qdrant.
“Top Matches” appear: e.g., “Maxime → Oscar (needs senior Java; Oscar is a Java backend dev).”
Organizer exports or shares intro cards/QRs.

Partner Tech Usage
Qdrant for vector search; Vercel for hosting; Firebase for auth/storage; Next.js/TypeScript for app; (Optional) Observable for fast data exploration/visualization during dev.

Technologies used
Frontend: Next.js, TypeScript
Backend/API: Next.js API routes (TypeScript)
Data & Auth: Firebase (Auth, Firestore/Storage)
Vector DB: Qdrant (similarity search, collections, payload filters)
Embeddings: provider-agnostic interface (plug-in for any text-embedding API)
Deployment/Hosting: Vercel
Dev/Exploration: Observable (for quick plots/inspection)
Autonomous agent logic (planned)
Multi-agent pipeline (Enrichment Agent, Entity-Resolution Agent, Matching Agent, Notifier Agent).
Policies: rate-limiters, exponential backoff, idempotent jobs, circuit breakers, fallbacks to rule-based matching when enrichment is sparse.
Human checkpoints for sensitive actions (intro emails, cross-org visibility).

Failure modes & mitigations
Sparse/Noisy Data → fall back to rule-based tags; prompt organizer to confirm key fields.
Enrichment/Rate-limit errors → queued retries, cached results, progressive disclosure.
Entity collisions (homonyms) → confidence scores + manual disambiguation UI.
False-positive matches → rationale + “why matched” features; organizer can down-rank/ban pairs to improve the model.
Privacy/Compliance → process only organizer-provided data + publicly available info; PII minimization and opt-out controls.

Next steps
Add autonomous agents for end-to-end orchestration and continuous enrichment.
Participant-facing experience (consent, preferences, opt-in icebreakers).
Real-time, on-site matching (scan badge → live suggestions).
Team-building and dating verticals (one high-confidence match/day).
Impact analytics (meetings booked, conversions, NPS).