📞 Closer AI Sales Call Evaluator

Overview

Closer AI Sales Call Evaluator is a robust, production-grade platform for analyzing sales call transcripts using OpenAI GPT-4, LangChain, and Supabase. It provides detailed, actionable feedback for sales teams, detects business rule violations at the chunk level, and supports advanced analytics, automation, and reporting.

---

✨ Key Features

• Automated Transcript Analysis: Processes and chunks transcripts, embedding each chunk for context-aware evaluation.
• Dynamic Business Rules: Enforce and update business rules (e.g., currency, compliance) at the chunk level, with violation detection and scoring penalties.
• Reference-Based Scoring: Compares each chunk to high-quality reference calls using vector search (Pinecone).
• Chunk-Level & Aggregated Reporting: Returns detailed JSON for each chunk and a comprehensive, aggregated final report.
• Async Analysis & Webhook Notification: Supports background processing and notifies external systems (e.g., Make.com) on completion.
• Supabase Integration: Stores all calls, analyses, business rules, and analytics in a scalable cloud database.
• Leaderboard & Analytics: Team and individual performance dashboards, coaching insights, and time-based metrics.
• Admin Controls: Secure authentication, call and rule management, and audit logging.
• Prompt Logging: All LLM prompts are logged for transparency and debugging.
• API-First: FastAPI backend with endpoints for all major operations.

---

🗂️ Project Structure

closer-ai-feedback-langchain/
├── api.py                  # FastAPI backend
├── main.py                 # CLI entry point for local analysis
├── database/
│   └── database_manager.py # Supabase integration and business logic
├── embeddings/
│   ├── embed_good_calls.py # Embedding reference calls
│   └── pinecone_store.py   # Pinecone vector store manager
├── langchain_script/
│   ├── analysis.py         # Core chunk/final analysis logic
│   ├── evaluator.py        # Transcript evaluation pipeline
│   └── transcript_parser.py# Transcript chunking utilities
├── data/
│   └── good_calls/         # Reference transcripts (not in repo)
├── requirements.txt
├── README.md
└── supabase_setup.sql      # Database schema

---

⚙️ Tech Stack

• OpenAI GPT-4 – LLM for deep analysis
• LangChain – RAG, prompt management, chunking
• Pinecone – Vector search for reference matching
• Supabase – Cloud PostgreSQL (with RLS, cascade, analytics)
• FastAPI – Secure, modern API
• Make.com – Workflow automation (webhook integration)
• Optional: React + Tailwind CSS frontend

---

🚀 Quickstart

1. Clone the Repository

git clone https://github.com/ren-418/closer-ai-feedback-langchain.git
cd closer-ai-feedback-langchain

2. Install Dependencies

python -m venv venv
source venv/bin/activate  # or venv\Scripts\activate on Windows
pip install -r requirements.txt

3. Configure Environment

Copy env_template.txt to .env and fill in your values:

# Required API Keys
OPENAI_API_KEY=your_openai_key
SUPABASE_URL=https://your-project.supabase.co
SUPABASE_KEY=your_supabase_key
PINECONE_API_KEY=your_pinecone_key
PINECONE_CLOUD=aws
PINECONE_REGION=us-east-1

# Make.com Webhook (for automation)
MAKE_WEBHOOK_URL=https://hook.us2.make.com/your_webhook_endpoint

4. Prepare Reference Data

• Place high-quality reference transcripts in data/good_calls/
• Run:

  python embeddings/embed_good_calls.py

5. Set Up Supabase

• Run supabase_setup.sql in your Supabase SQL editor to create tables and policies.
• If you already have a database, ensure DELETE policies are present for RLS (see add_delete_policies.sql if needed).

6. Run the API Server

uvicorn api:app --reload
python api.py

7. Analyze a Call (CLI)

python main.py path/to/transcript.txt

---

🛡️ Security & Data Privacy

• RLS: Row Level Security is enabled on all tables.
• No real transcripts are included in the repo.
• All prompts and analyses are logged for auditability.

---

🧠 Advanced Capabilities

• Business Rules Engine: Add, update, or remove rules via API. Violations are detected per chunk and aggregated in reports.
• Chunk-Level Analysis: Each transcript is split and analyzed in context, with reference matching and business rule enforcement.
• Async Processing: New calls can be submitted for background analysis, with webhook notification on completion.
• Comprehensive Analytics: Leaderboards, coaching insights, and time-based performance metrics for teams and individuals.
• Admin & API Controls: Secure endpoints for user, closer, call, and rule management.

---

📈 Example Use Cases

• Sales Team Coaching: Identify strengths, weaknesses, and compliance issues in real calls.
• QA & Compliance: Enforce business rules (e.g., currency, legal language) and detect violations in real time.
• Automated Reporting: Integrate with Make.com or other tools for workflow automation and notifications.

---

🛠️ Roadmap

• Dynamic business rules (API-managed)
• Chunk-level violation detection
• Async analysis & webhook notification
• Analytics dashboard (API)
• React frontend (optional)
• More granular RLS policies

---

📄 License

Private project — Not open-sourced. Usage permitted only by client or authorized team members.