Converge is an AI-powered personal CRM that captures the essence of real conversations at networking events—so you can build authentic, lasting relationships instead of forgetting context 48 hours later.

• Demo video: (add link)
• Devpost: (add link)

Traditional networking tools capture transactional data (name/company), but miss what actually matters:

• what you discussed
• what they needed
• what you promised
• where you met
• what they looked like

Converge turns a fleeting conversation into durable, queryable memory.

• 🔐 Authentication (email/password, JWT sessions)
• 🎥 Record conversations with event context (audio + video)
• 🎧 LiveKit audio processing
  • transcription
  • structured profile extraction (name/company/role/topics/challenges/follow-up hooks)
  • confidence scores (high/medium/low)
• 🎞️ Overshoot video processing
  • face tracking + 512-d face embedding
  • appearance + environment description
• 🧩 Merge + store audio + visual into a single draft profile in MongoDB Atlas
• 📝 Approval UI
  • highlights low-confidence fields
  • inline edits
  • approve / discard
• 🗂️ Connections grid + detail view
  • browse and search your network
  • view follow-up hooks and conversation context
• 🎙️ Voice agent search (“Who was the scout from a16z?”)
• 📈 Analytics dashboard (connections over time, follow-ups pending, top events)
• 💌 Smart follow-up suggestions (personalized messages based on real conversation)
• 🕸️ Network graph visualization

flowchart LR
  A[React Frontend] <-- REST / WS --> B[Express Backend]
  B <--> C[MongoDB Atlas]
  A <-- WS Audio --> D[LiveKit]
  A <-- WS Video --> E[Overshoot]
  B --> F[ElevenLabs TTS (Stretch)]
  C --> G[Atlas Vector Search<br/>face_embedding (512)]
Data Flow (Capture → Process → Merge → Approve → Use)
Capture: device records audio + video

Process (parallel):

LiveKit → transcript + profile fields + confidence

Overshoot → face embedding + appearance + environment

Merge: backend merges outputs + context into a draft profile

Approve: user reviews/edits low-confidence fields → approve

Use: browse/search network, follow up (stretch: voice search + suggestions)

Tech Stack 🧰 Frontend: React (Vite recommended)

Backend: Node.js + Express

DB: MongoDB Atlas (+ Atlas Vector Search)

Audio: LiveKit (WebSocket streaming + transcription/extraction)

Video: Overshoot (WebSocket streaming + embeddings + descriptions)

TTS (stretch): ElevenLabs

LLM provider: Anthropic

Project Structure 📁 Adjust paths to match your repo. This is the recommended layout.

bash Copy code converge/ frontend/ # React app (Vite) src/ package.json

converge-mobile/ # Mobile React app (Vite) client/ src/ package.json

backend/ # Express API + WS handlers src/ routes/ middleware/ services/ ws/ # LiveKit/Overshoot WS gateways (if proxied) db/ package.json

docs/ # PRD, diagrams, demo assets README.md Quickstart (Local Dev) ⚡ 0) Prerequisites Node.js 18+

MongoDB Atlas account + cluster

LiveKit credentials (or mocked test payloads)

Overshoot credentials (or mocked test payloads)

1. Clone + install bash Copy code git clone cd converge

cd backend npm install

cd ../frontend npm install 2) Configure environment variables Create .env files from the examples:

backend/.env bash Copy code

MONGODB_URI=mongodb+srv://...

JWT_SECRET=your-secret-key

LIVEKIT_API_KEY=... LIVEKIT_API_SECRET=... LIVEKIT_URL=wss://...

OVERSHOOT_API_KEY=... OVERSHOOT_WS_URL=wss://...

ELEVEN_LABS_API_KEY=... ELEVEN_LABS_VOICE_ID=...

PROXYCURL_API_KEY=...

ANTHROPIC_API_KEY=... frontend/.env bash Copy code

VITE_API_BASE_URL=http://localhost:5000

Auth Method Endpoint Description POST /api/auth/signup Create account POST /api/auth/login Login + JWT POST /api/auth/logout Logout GET /api/auth/me Current user

Connections (MVP) Method Endpoint Description POST /api/connections/merge Merge LiveKit + Overshoot outputs into draft GET /api/connections List connections (supports search/filter/pagination) GET /api/connections/:id Get a connection PATCH /api/connections/:id/approve Approve draft with edits DELETE /api/connections/:id Discard draft

Search (Stretch) Method Endpoint Description POST /api/search/voice Voice transcript → parsed query → results (+ TTS) POST /api/search/text Text query fallback

Analytics (Stretch) Method Endpoint Description GET /api/analytics/overview Totals + key metrics GET /api/analytics/industry-breakdown Group by industry GET /api/analytics/connections-over-time Trend line GET /api/analytics/top-events Events ranked GET /api/analytics/relationship-types Breakdown

Database Schema (High-Level) 🗄️ users email (unique), password_hash, name

settings (default_event, voice_agent_enabled)

connections status: draft | approved | archived

identity: name/company/role (value + confidence + source)

visual: face_embedding (512), appearance, environment

audio: topics, challenges, follow_up_hooks, transcript_summary

context: event, location, first_met

review flags: needs_review, fields_needing_review

interactions logs meetings/follow-ups with timestamps + summary snapshots