Agent Transponder

Automated meeting-to-CRM pipeline with AI-powered follow-up notes.

Agent Transponder automatically processes your video call recordings, extracts actionable CRM notes using AI, uploads them to HubSpot, and emails you a professional follow-up summary — all without lifting a finger.

---

What It Does

┌──────────────┐     ┌──────────────────┐     ┌─────────────┐     ┌─────────────┐
│   Fathom     │────▶│  Fathom CRM      │────▶│  HubSpot    │────▶│   Email     │
│  (Meeting)   │     │  Agent (AI)      │     │  (CRM Note) │     │  (Summary)  │
└──────────────┘     └──────────────────┘     └─────────────┘     └─────────────┘
     Webhook              Claude/Groq LLM              REST API           Resend API

1. You finish a meeting recorded by Fathom
2. Fathom sends a webhook with transcript, summary, and action items
3. AI extracts company name, attendees, CRM-ready notes, and next steps
4. HubSpot receives a formatted note attached to the correct company
5. You receive an email with the follow-up summary and a link to the CRM record

---

Features

• Zero manual data entry — meetings automatically sync to your CRM
• AI-powered extraction — Claude (claude-sonnet-4-20250514) identifies key information
• Smart company matching — finds the right HubSpot company by domain or name
• Professional email summaries — formatted follow-up notes delivered to your inbox
• Audit trail — all processed meetings logged in PostgreSQL

---

Tech Stack

Component	Technology	Purpose
Backend	FastAPI (Python 3.11+)	Webhook handling, API routing
Database	PostgreSQL (Neon)	User accounts, credentials, meeting logs
AI	Claude API / Groq API (Llama 3.1 70B)	Transcript processing, note extraction
CRM	HubSpot API	Company search, note creation
Email	Resend	Follow-up email delivery
Hosting	Railway / Render	Web service deployment

---

Quick Start

Prerequisites

• Python 3.11+
• PostgreSQL database (Neon free tier works)
• API keys for: Fathom, HubSpot, Claude / Groq, Resend

1. Clone and Install

git clone https://github.com/Shiv716/Agent-Transponder.git
cd agent-transponder
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

2. Configure Environment

cp .env .env
# Edit .env with your API keys

3. Initialize Database

python -m app.core.database --init

4. Run Locally

uvicorn app.main:app --reload --port 8000

5. Register Fathom Webhook

Point your Fathom webhook to:

https://your-domain.com/webhook/fathom

---

Environment Variables

Variable	Description	Required
DATABASE_URL	PostgreSQL connection string	✅
LLM_API_KEY	LLM API key for LLM processing	✅
HUBSPOT_ACCESS_TOKEN	HubSpot private app access token	✅
RESEND_API_KEY	Resend API key for emails	✅
FATHOM_WEBHOOK_SECRET	Fathom webhook signing secret	✅
USER_EMAIL	Email address for follow-up delivery	✅
APP_ENV	development or production	❌

---

API Endpoints

Method	Endpoint	Description
POST	/webhook/fathom	Receives Fathom meeting webhooks
GET	/health	Health check for monitoring
GET	/meetings	List processed meetings
GET	/meetings/{id}	Get specific meeting details

---

Project Structure

agent-transponder/
├── app/
│   ├── main.py              # FastAPI application entry
│   ├── core/
│   │   ├── config.py        # Environment configuration
│   │   └── database.py      # Database connection & models
│   ├── models/
│   │   └── schemas.py       # Pydantic models
│   ├── routers/
│   │   ├── webhook.py       # Fathom webhook handler
│   │   └── meetings.py      # Meeting history endpoints
│   └── services/
│       ├── ai_processor.py  # LLM integration
│       ├── hubspot.py       # HubSpot API client
│       └── email.py         # Resend email service
├── tests/
│   └── test_webhook.py      # Webhook handler tests
├── docs/
│   ├── SETUP.md             # Detailed setup guide
│   └── ARCHITECTURE.md      # System design documentation
├── .env.example             # Environment template
├── requirements.txt         # Python dependencies
├── render.yaml              # Render deployment config
├── CONTRIBUTING.md 
└── README.md

---

Deployment (Render)

One-Click Deploy

Manual Deploy

1. Create a new Web Service on Render
2. Connect your GitHub repository
3. Set environment variables in Render dashboard
4. Deploy

The included render.yaml handles build and start commands automatically.

---

How AI Processing Works

The LLM receives the meeting transcript and extracts:

{
  "company_name": "Acme Corp",
  "company_domain": "acme.com",
  "attendees": ["John Smith", "Jane Doe"],
  "crm_note": "Discussed Q3 implementation timeline. Client confirmed budget approval. Next steps: send SOW by Friday.",
  "action_items": [
    "Send SOW document",
    "Schedule technical deep-dive",
    "Confirm stakeholder list"
  ],
  "meeting_sentiment": "positive",
  "deal_stage_signal": "negotiation"
}

This structured data drives the HubSpot update and email content.

---

License

MIT

---

Contributing

Contributions welcome! Please read CONTRIBUTING.md for guidelines.

---

Support

• Issues: GitHub Issues
• Documentation: docs/

---

Built with ☕ and AI.