A multi-agent automation system that monitors business opportunities in real-time, evaluates fit using NLP, and auto-generates personalized outreach — reducing manual work by over 90%.
Built as an internal tool to automate a high-volume, repetitive business development workflow. The system continuously monitors opportunity feeds via REST API polling, classifies each opportunity using a hybrid NLP pipeline (rule-based filters + LLM feasibility scoring via Claude API), and generates context-aware personalized responses — all without human intervention.
State is persisted atomically (tmp file then rename) to prevent corruption during crashes. A Telegram bot interface gives real-time visibility and control: notifications, status updates, manual overrides, and performance reporting. Over 2,800 opportunities have been processed to date.
The system runs as a persistent background service on a Linux VPS. Core components: an opportunity fetcher (REST API polling), a classifier (rule-based + LLM hybrid), a response generator (Claude API), and a Telegram bot for operator control. State is persisted in JSON with atomic writes to prevent corruption.