LIVE2025

LeadGen Pro — AI-Powered B2B Lead Generation Engine

LeadGen Pro is a self-hosted B2B lead generation engine designed for zero marginal cost at any volume. It continuously scrapes GitHub, GitLab, HackerNews "Who wants to be hired?", Google Custom Search, and custom URLs using 80+ built-in query templates across India, USA, UK, Europe, APAC, LATAM, and the Middle East. Each lead is deduplicated via SHA-256 email fingerprinting and Levenshtein fuzzy matching on name + domain, then enriched by Mistral 7B running locally on a GTX 1050 GPU via CUDA — identifying pain points, scoring quality (hot/warm/cold), assigning a confidence score (0–100), and generating hyper-personalised outreach copy referencing the lead's specific role and company. A daily target mode runs the pipeline continuously until your quota is hit, auto-pauses until midnight, and resumes. An hourly sweep catches any leads Ollama missed during the main run. The entire stack is production-hardened: JWT-authenticated API, rate limiting on all endpoints, CentOS Stream 9, Nginx reverse proxy, PM2 process manager, and DKMS NVIDIA driver signed for Secure Boot.

Launch Application

$ Architecture

→Frontend: React 18 + Vite + TailwindCSS + TanStack Query. Real-time dashboard with live WebSocket updates via Socket.io. Filter by quality, source, status, date, location. Manual category override and per-lead notes.
→Backend: Node.js 20 (ESM) + Express.js. JWT HS256 auth, rate limiting on all endpoints. REST API for lead management, scraping control, AI enrichment triggers, and CSV export.
→AI Layer: Mistral 7B running locally via Ollama on GTX 1050 GPU (CUDA). Two-pass enrichment — first pass: pain point identification, quality scoring (hot/warm/cold), confidence score (0–100). Second pass: hyper-personalised outreach copy guided by ICP + product description from settings.
→Scraping Engine: Axios + Cheerio for HTML sources, GitHub API, Algolia HN API, Google Custom Search Engine. Rate-limited with randomised user-agents and per-domain throttling. 80+ built-in GitHub query templates covering global regions.
→Deduplication: SHA-256 email fingerprinting for exact match, Levenshtein fuzzy match on name + domain to catch variations. Runs before AI enrichment to avoid wasting GPU cycles.
→Pipeline Orchestration: Daily target mode (e.g. 500/day) runs back-to-back scrape → dedup → enrich cycles with zero idle time. Auto-pauses on quota hit, resumes at midnight. Hourly sweep catches leads missed during main run.
→Infra: CentOS Stream 9, Nginx reverse proxy, PM2 process manager. DKMS NVIDIA driver signed for Secure Boot. SQLite (zero-config) or PostgreSQL — swappable via env var.

$ Tech Stack

Node.jsExpress.jsReactViteTailwind CSSTanStack QuerySocket.ioMistral 7BOllamaNVIDIA CUDASQLitePostgreSQLAxiosCheerioJWTNginxPM2DockerCentOS Stream 9

$ Outcomes

→Automated multi-source scraping: GitHub, GitLab, HackerNews, Google CSE, and custom URLs with 80+ built-in query templates across global regions

→100% local AI enrichment via Mistral 7B on Ollama — zero API cost, zero data leaving your server

→Two-pass AI pipeline: lead qualification + scoring on first pass, hyper-personalised outreach copy on second pass

→Smart deduplication: SHA-256 email fingerprinting + Levenshtein fuzzy match on name + domain

→Real-time dashboard with live WebSocket updates — filter by quality, source, status, date, and location

→Manual category override (hot/warm/cold/disqualified) and per-lead notes for human-in-the-loop refinement

→Daily target mode: set a lead quota (e.g. 500/day), pipeline runs continuously until hit, auto-pauses until midnight

→Continuous pipeline with hourly sweep catching any leads Ollama missed during the main run

→CSV export for dropping leads into any CRM or outreach tool

→Configurable ICP + product description — every future lead scored against your exact criteria

→JWT-authenticated API with rate limiting on all endpoints

→Zero recurring AI cost — no OpenAI/Claude/SaaS bills; limited only by hardware and scraping rate limits