RFP Harvester
The system that finds Qatalyst's bidding opportunities. It checks more than 40 procurement sources three times a day, removes duplicates and expired notices, and has AI score each one against the firm's service lines.
- Python
- BeautifulSoup
- Selenium
- Anthropic SDK
- Supabase
- Langfuse
- Docker
- Railway
The problem
A consulting firm depends on finding the right requests for proposals in time. Public opportunities are scattered across dozens of federal, provincial, territorial and international portals, and most of them aren’t a fit.
What I built
- A harvester that runs three times a day across more than 40 procurement sources in Canada and internationally, including MERX, CanadaBuys, BC Bid, provincial and territorial portals, the World Bank and TED Europa.
- Clean-up. Duplicates and expired notices are removed.
- AI fit scoring against Qatalyst’s service lines and past work.
- Assessment. Staff review the scored opportunities and run structured fit assessments. Associates can express interest, and AI-written fit summaries help staff choose the team.
Under the hood
Python, with an email-alert parser or web scraper for each source, running as a scheduled Docker service. Scoring uses Claude, traced in Langfuse. Results land in Qatalink, where the rest of the bid process happens.








