All work

AI agent · Professional work

Tender Scraper Agent

Give it a tender website and it explores the site, writes a scraper, tests it and fixes its own errors until it works. Used on 280+ real government tender websites.

TENDER EXPLORERFind. Extract. Repeat.$ agent.run(website)↳ exploring tender list↳ writing recipe↳ testing & repairingrecipe ready for approval

The problem

A tender-aggregation platform tracks hundreds of government and bank tender sites. Each needed a hand-written scraper, and sites change often.

What I built

An agent that, given a URL, surveys the site, finds the tender list, chooses fields (title, deadline, value, documents), writes a scraping recipe, runs it on the real scraping engine and repairs it from structured errors. Admins approve with one click.

How it works

  1. Explore — survey the website and locate the tender list, detail pages and documents.
  2. Write — choose the fields and produce a scraping recipe for the existing engine.
  3. Test and repair — run the recipe on the real scraping engine, read structured errors and fix the recipe within the job budget.
  4. Approve — hand the working recipe to an admin for one-click approval.

The hard parts

Pagination, detail pages, document downloads, pop-ups and many date formats all needed handling without site-specific code. Image CAPTCHAs use an AI solver with a human fallback. Every job has a 2-hour budget.

The result

Shipped to the team's dev environment across 3 codebases; QA uses it on 280+ tender websites. The agent produced working recipes for the majority of sites without human help.

Built with

  • TypeScript
  • LLM agents
  • MCP tools
  • browser automation

Professional work. This write-up describes the workflow; source code and internal data are private.

Next project

Slack HR Onboarding Agent

Let’s make something useful

Have a project in mind?

Tell me about it. I reply within 24 hours.