🛰️ B73 Lead Radar

AI-assisted lead monitoring and qualification pipeline built for B73 Digital Studio.

It collects structured opportunities, stores them in SQLite, evaluates commercial intent using NVIDIA Nemotron, and surfaces qualified leads through Telegram.

---

✨ What It Does

SOURCE
  ↓
COLLECT
  ↓
NORMALIZE
  ↓
DEDUPLICATE
  ↓
SQLITE
  ↓
AI QUALIFICATION
  ↓
TELEGRAM

The system is designed to separate actual buyer intent from low-value or irrelevant opportunities.

⚙️ Core Features

• Python-based modular pipeline
• SQLite persistence
• Duplicate-safe ingestion
• NVIDIA NIM integration
• Nemotron-powered lead qualification
• Telegram operator bot
• Manual lead ingestion
• Experimental EU TED procurement collector
• Environment-based secret management
• Modular collector architecture

🧠 AI Qualification

Default model: nvidia/nemotron-3.5-lightning-30b-a3b

Optional deeper model: nvidia/nemotron-3-ultra-550b-a55b

The qualifier evaluates signals such as:

• buyer intent

• commercial urgency

• scope clarity

• fixed-price suitability

• budget signals

• relevance to web/software development

  🧱 Architecture

            ┌──────────────┐
            │  Data Source │
            └──────┬───────┘
                   ↓
            ┌──────────────┐
            │  Collector   │
            └──────┬───────┘
                   ↓
            ┌──────────────┐
            │ Normalizer   │
            └──────┬───────┘
                   ↓
            ┌──────────────┐
            │ Deduplication│
            └──────┬───────┘
                   ↓
            ┌──────────────┐
            │   SQLite     │
            └──────┬───────┘
                   ↓
            ┌──────────────┐
            │  Nemotron AI │
            └──────┬───────┘
                   ↓
            ┌──────────────┐
            │   Telegram   │
            └──────────────┘
  

🛰️ Telegram Commands

/start

/status

/analyze_new

🌍 EU TED Collector

The repository includes an experimental collector for the official EU TED procurement search API. Current implementation demonstrates:

• official API integration
• query construction
• response parsing
• normalization into the internal Lead model
• SQLite ingestion
• duplicate-safe storage

The TED collector is included as a reference implementation and may require query tuning depending on the search scope and CPV filters.

🧪 Demo Lead

A demo lead can be added with: python seed_test_lead.py

Then analyze it through Telegram: /analyze_new

🚀 Installation

Clone the repository:

git clone https://github.com/fatihex3/b73-lead-radar.git
cd b73-lead-radar

Create a virtual environment:

python3 -m venv venv
source venv/bin/activate

Install dependencies:

pip install -r requirements.txt

Create your environment file:

cp .env.example .env

Add your own credentials to .env.

Run:

python main.py

🔐 Security

Never commit:

• .env
• API keys
• Telegram tokens
• SQLite production databases
• SSH keys
• production logs This repository contains no production credentials.

🛠️ Tech Stack

• Python
• SQLite
• Telegram Bot API
• NVIDIA NIM
• Nemotron
• HTTPX
• OpenAI-compatible API client

📌 Project Status

Public portfolio/reference release. The production version contains additional reliability, observability, qualification and ingestion safeguards.

👨‍💻 Built by

B73 Digital Studio

https://b73.dev