Business Case: Finding and qualifying government contracts across 50 different state portals is a massive resource drain for contractors. Legacy scrapers fail because portals like Texas SmartBuy and Florida Marketplace use complex search grids, date-picker popups, and nested document tables. TenderFlow eliminates this bottleneck. By deploying a dynamic AI swarm, it navigates complex UI workflows like a human operator, captures precise solicitation scopes, and extracts verified bid documents in seconds. It transforms procurement from a manual data-entry job into an automated, high-velocity intelligence feed.

Technical Architecture: TenderFlow is built on a custom Node.js/Express orchestration backend integrated with the TinyFish API.

Sequential Compute Engine: To maximize a single-node concurrency limit, the backend utilizes an async polling queue that strictly manages node hand-offs between separate "Scout" (search) and "Worker" (extraction) agents.

Deterministic Extraction Protocol: To prevent URL hallucination, the LLM prompts mandate raw HTML DOM inspection to capture underlying authenticated media.nl?id= strings rather than guessing file paths.

Live Telemetry Pipeline: The frontend UI uses Server-Sent Events (SSE) to consume real-time agent step-logs and pipes the active TinyFish streaming_url directly into native iFrames, providing zero-latency system transparency to the user

Built With

Share this project:

Updates