# zhalpakov

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Aristid Popandopulo](https://cerebralvalley.ai/u/Futurizm), [Baraat Kahimalaev](https://cerebralvalley.ai/u/Baraat04)
- **GitHub:** https://github.com/Baraat04/xscan
- **Demo video:** https://youtu.be/BlgJTwSCjxY?si=J90OX0pEN-2wrSW5
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/493

xscan: Tender Tribunal

Live demo: https://xscan.baby (https://xscan.baby/)

One line: xscan puts suspicious public tenders on trial, with AI agents arguing both sides and every quote checked against the original notice.

The problem

Governments spend huge sums through public tenders, and a contract can be quietly steered to one supplier: a brand name with no “or equivalent”, a deadline too short for anyone else to bid, or a turnover requirement far above the contract value. These signs sit inside long notices in many languages, and few people have time to read them. A single AI summary doesn’t solve this, because it can sound confident, invent evidence and ignore the buyer’s side.

What xscan does

xscan collects live tenders from the EU (TED), the US (USAspending, and SAM.gov (http://sam.gov/) with an API key) and Kazakhstan (zakup.gov.kz (http://zakup.gov.kz/)). Users can also add a tender or a whole listing from a procurement website by pasting a link.

Thirteen free rules screen every tender first, so AI is only spent on the ones with warning signs. A flagged tender then goes to a hearing run by Meta’s Muse Spark model:

The Prosecutor finds conditions that seem tailored to one supplier and must quote the notice word for word for each count.
The Defense gives the buyer’s most honest legitimate explanation, and concedes when there is none.
The Court rules count by count, sets a risk score and names possibly relevant rules, marked for verification.
The Clerk drafts a formal request for clarification to the buyer, in the notice’s own language, with an English version.
Why you can trust the output

Quotes are verified: every quote is checked against the notice text, in code rather than by the model. A quote that isn’t in the notice is struck through and given no weight.
Tender text is treated as untrusted data: a notice that tries to instruct the AI is flagged as a warning sign instead of being obeyed.
The record is auditable: each hearing records the model, prompt version, latency and tokens used.
People make the final call: xscan points out warning signs for human review and never calls a tender corrupt.
How it’s built

Frontend: Next.js and React, with hearings streamed live to the page, so a reload picks up where it left off.
Backend: Node.js, Express and PostgreSQL.
AI: Muse Spark 1.3 on the Meta Model API, using JSON-schema structured output, validation and bounded retries.
Reading websites: headless Chromium opens JavaScript-heavy portals, and every request is checked so a page can’t reach internal addresses.
Deployment: Docker with automatic HTTPS.
Who it’s for

Journalists, watchdog NGOs, auditors, compliance teams and suppliers who suspect a tender was written for someone else.

## More from Meta Global AI Developer Hackathon

- [ZionMatrix - ASO](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/490)
- [Swastik Mishra](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/491)
- [KINETIC-ORACLE](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/492)
- [Kuron Universe](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/494)
- [Savy Electrical](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/495)
- [Plumb](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/496)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/493. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt