# Postor

- **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:** [Aneudys Amparo](https://cerebralvalley.ai/u/Aneudys)
- **GitHub:** https://github.com/aneudysamparo/Postor-MetaAI-Hackathon
- **Demo video:** https://youtu.be/-CQuarEifQY
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1971

Postor AI Bid Team takes a small business from a real public tender to an honest answer: GO, GO WITH CONDITIONS, NO GO or NEEDS REVIEW, plus a bid-ready draft when bidding makes sense.

---
For judges: open the live demo at https://demo.postor.do and, on the sign-in page, enter the access code ("3dee7c965d882447")  and your own Meta Model API key. The key powers the live analyses: it stays in your browser tab, travels only with your requests to Meta and is never stored by our server. 
Suggested 3-minute path: Opportunities → open Case A → ▶ Run live analysis (Fast mode, ~2 min) → Analysis (verdict, conditions, financial ratios) → Requirements (click any "pliego pN" to see the cited page of the original PDF) → Trace (every Meta model call with prompt, response, tokens and estimated cost). 
Then compare Cases C and D: same pipeline, different verdicts, each for a cited reason. The four tenders also include recorded runs to browse while a live analysis runs. The README has the full walkthrough.
---

The problem: so far in 2026, the Dominican state has awarded more than RD$437 billion in public contracts (RD$324B in 2025, RD$191B in 2024). Every peso goes through a tender that a small business has to read page by page, and bids are lost on paperwork: a missing certificate, an expired document, a requirement found too late.

Postor is a live, paid SaaS at postor.do with a small base of paying customers. Before this hackathon it helped companies find tenders and see what each one asks for. With Meta's models we built the rest of the path:
- Reads the real tender (pliego, amendments, official forms), never modified, every page screened for prompt injection.
- Checks each requirement against the company's own evidence (certifications with expiry dates, projects, finances, team), citing document + page on both sides.
- Decides with deterministic guards: financial ratios, revenue concentration and mandatory pass/fail criteria are code, not model output.
- Drafts the technical proposal, has a second model (Muse Wonder) audit it, fills the tender's own forms and prices the F.033 economic offer at ~97% of the budget, tax included, with the total in words.
- Leaves the human in charge: anything unproven stays [PENDING], and the bid manager can answer by voice. Postor never submits, signs or sends.

On 4 real Dominican tenders the same pipeline says go, no go or needs review for the right reasons; a contract worth 86% of the company's revenue goes to a human. Every model call is traceable (prompt, response, tokens, duration, estimated cost). The demo company is fictional and marked DEMO; the tenders are real public records.

Video: https://youtu.be/-CQuarEifQY 
X: https://x.com/AneudysAmparo/status/2117991038019396050

## More from Meta Global AI Developer Hackathon

- [ApplyArc](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1968)
- [Magister AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1969)
- [AfterBooth](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1970)
- [PHISHGUARD AI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1972)
- [Md sabbir sikder](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1973)
- [continuum](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1974)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1971. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt