DEALROOM

Real-Time Negotiation Intelligence Agent

Your silent AI copilot for every sales call

Gemini Live Agent Challenge | March 2026

Project Overview

• Project: DealRoom
• Tagline: Your silent AI copilot for every sales call
• Challenge: Gemini Live Agent Challenge
• Repository: https://github.com/1AyaNabil1/DealRoom

Hackathon Requirement Coverage (Judge Checklist)

This section maps each mandatory requirement to concrete evidence in this repository.

1. Leverages a Gemini model

• Evidence: Google GenAI SDK usage in src/agent.py and src/context_merger.py
• Evidence: Gemini model usage with live session connect in src/agent.py

2. Built with Google GenAI SDK or ADK

• Evidence: Google GenAI SDK imports and client/session usage in src/agent.py and src/context_merger.py

3. Uses at least one Google Cloud service

• Evidence: Google Cloud Text-to-Speech API integration in src/server.py
• Evidence: Vertex AI endpoint API call example in src/gcp_vertex_demo.py
• Direct GitHub link (requirement option 2 proof): https://github.com/1AyaNabil1/DealRoom/blob/main/src/gcp_vertex_demo.py

4. Public code repository with reproducible spin-up instructions

• Evidence: This README Quick Start section

5. Google Cloud deployment proof

• Evidence (infrastructure and deployment code): infra/main.tf, Dockerfile, scripts/deploy.sh
• Requirement-compatible proof link to Google Cloud API usage code: https://github.com/1AyaNabil1/DealRoom/blob/main/src/gcp_vertex_demo.py

6. Architecture diagram

• Evidence: Architecture/data flow section below (and diagram image should be uploaded in Devpost media carousel)

Text Description

DealRoom is a real-time negotiation copilot that runs as a floating overlay during sales calls. It captures microphone audio in short windows, sends context to an AI agent, classifies live signals (TACTIC, SIGNAL, RED_FLAG), and returns concise coaching cards. At session end, it generates a structured debrief and can read critical guidance aloud.

Core Features

• Real-time tactical coaching cards over WebSocket
• Red-flag detection for risk moments
• Post-call debrief generation
• Text-to-speech playback with layered fallback
• Floating overlay UI that stays visible but non-blocking

Technologies Used

• Backend: FastAPI + WebSocket + asyncio
• AI SDK: Google GenAI SDK
• Google Cloud APIs: Text-to-Speech, Vertex AI endpoint example
• Frontend: Vanilla JavaScript + MediaRecorder
• Deployment assets: Dockerfile + Terraform

Quick Start (Spin-Up Instructions)

1. Install dependencies

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

2. Configure environment

cat > .env << 'EOF'
GOOGLE_API_KEY=your_google_ai_studio_key
# Optional for Google Cloud TTS
GCP_PROJECT_ID=your_gcp_project_id
# Optional for Vertex endpoint demo
GCP_LOCATION=us-central1
VERTEX_ENDPOINT_ID=your_vertex_endpoint_id
EOF

3. Run backend

uvicorn src.server:app --host 0.0.0.0 --port 8080 --env-file .env

4. Open overlay

Go to http://localhost:8080/overlay and allow microphone access.

Google Cloud API Proof (Code Links)

• Vertex AI endpoint call demo file: src/gcp_vertex_demo.py
• Google Cloud TTS request flow: src/server.py
• Terraform enabling GCP services including Vertex AI and Cloud Run: infra/main.tf

Architecture Summary

1. Browser overlay captures microphone chunks.
2. Chunks stream to FastAPI over WebSocket.
3. Backend sends contextual prompts to the AI engine and parses structured JSON responses.
4. Overlay renders TACTIC/SIGNAL/RED_FLAG cards.
5. TTS endpoint synthesizes key messages via Google Cloud TTS (with fallback path).
6. Debrief endpoint returns a compact post-session summary.

Bonus Criteria Coverage

• Infrastructure as code: infra/main.tf
• Containerized deployment path: Dockerfile
• Deployment script: scripts/deploy.sh

---

Built for the Gemini Live Agent Challenge.