Built a full-stack lead discovery platform that automatically monitors the web and local business directories for prospects who could benefit from life, business, or systemic coaching
Engineered a dual-source data pipeline: Google Alerts RSS (continuous web monitoring every 20 minutes) and Google Places API (daily business review analysis), replacing manual prospecting entirely
Integrated a local LLM (Ollama/Llama3) to score and analyze leads across four dimensions — pain signals, growth signals, structural need, and personal strain — producing a composite 0–100 score and coaching fit rating (HIGH/MED/LOW)
Designed a duplicate-detection system using MD5 content fingerprinting and source URL deduplication to prevent redundant lead storage
Built a bilingual analysis pipeline (English and Romanian) with keyword-based pre-filtering to optimize LLM inference cost and latency
Implemented automated email alerts (hot and warm leads) and daily digest summaries via Nodemailer with local Postfix relay
Developed a session-based authentication system with MySQL-persisted sessions, cookie and header token support, and automatic session expiry
Built an outreach message generator that produces personalized, non-salesy contact messages via the local LLM, respecting language and tone preferences
Deployed behind Nginx reverse proxy as a systemd service on a VPS with IPv4-forced HTTP agents to ensure network reliability
WebBetas
