Skip to content

resume_agent

Directory actions

More options

Directory actions

More options

Latest commit

 

History

History

resume_agent

README.md

Voice Agent: Personal Recruiter

A phone-callable AI agent that represents your professional profile to recruiters. Built with Syrin (agent brain) + Pipecat (voice pipeline).

Setup

cd examples/resume_agent
uv venv .venv
uv sync
source .venv/bin/activate  # Windows: .venv\Scripts\activate

Architecture

See ARCHITECTURE.md for the full system diagram.

Flow: Twilio (phone) → Pipecat (STT/TTS/VAD) → Syrin Agent (Knowledge, tools) → response

Quick Start (Text Mode)

From this directory with the venv activated:

source .venv/bin/activate  # or .venv\Scripts\activate on Windows
python agent.py

Type questions; the agent uses the knowledge base (data/about.md, data/skills.yaml, data/projects/).

Voice Mode (Phone Call)

1. Install Pipecat

pip install "pipecat-ai[deepgram,elevenlabs]"
# Or: uv sync (from this directory)

2. Environment

Ensure examples/.env (parent of resume_agent) has:

  • OPENAI_API_KEY — LLM + embeddings
  • DEEPGRAM_AUTH_TOKEN or DEEPGRAM_API_KEY — STT
  • ELEVENLABS_API_KEY — TTS
  • TWILIO_ACCOUNT_SID, TWILIO_AUTH_TOKEN — telephony

3. Customize Your Data

Edit data/about.md, data/skills.yaml, and data/projects/ with your resume and projects. The agent answers from this knowledge.

4. Run Locally with ngrok

# Terminal 1: Expose local server
ngrok http 7860

# Terminal 2: Start voice server (from examples/resume_agent/)
source .venv/bin/activate
python voice_server.py -t twilio -x YOUR_NGROK_SUBDOMAIN.ngrok.io

Use -x with your ngrok host (e.g. abc123.ngrok-free.app) so Twilio can reach the WebSocket.

5. Configure Twilio

  1. Go to Twilio Console → Phone Numbers → your number.
  2. Under "Voice Configuration", set "A call comes in" to TwiML Bin.
  3. Create a TwiML Bin with:
<?xml version="1.0" encoding="UTF-8"?>
<Response>
  <Connect>
    <Stream url="wss://https://6487-2405-201-d061-7-109e-2475-3a67-5aeb.ngrok-free.app/ws" />
  </Connect>
</Response>

Replace YOUR_NGROK_HOST with your ngrok URL (e.g. abc123.ngrok-free.app).

6. Call

Call your Twilio number. The agent will greet you and answer questions about the resume content in data/.

Optional: Postgres Knowledge Backend

For production, use Postgres + pgvector for Knowledge:

  1. Start Postgres with pgvector:
    docker run -d --name syrin-pg -p 5434:5432 -e POSTGRES_PASSWORD=postgres pgvector/pgvector:pg16

  2. Set in examples/.env:

    DATABASE_URL=postgresql://postgres:postgres@localhost:5434/postgres

  3. Install: uv sync --extra knowledge-postgres

The agent will use Postgres for the knowledge store.

Configuration You May Need

Need Action
Different ElevenLabs voice Set ELEVENLABS_VOICE_ID in .env
Deepgram key Use DEEPGRAM_AUTH_TOKEN (mapped to DEEPGRAM_API_KEY in code)
Twilio number Ensure it supports Voice; configure the TwiML Bin as above