# Mohd Azhar Mansoori

- **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:** [Mohd Azhar](https://cerebralvalley.ai/u/Sider0024)
- **GitHub:** https://github.com/azhardev77/tenderops-ai
- **Demo video:** https://youtu.be/9jeeBAneTFM?si=gdn86HzwQ7AtiydF
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5279

TenderOps AI reads government and enterprise tender PDFs and answers one question: can this company bid, and why? Checking a 100-page tender against company records by hand is slow and error-prone, and one missed clause means a rejected bid.

How it works: (1) SAM 3.1 locates stamps, signatures and tables on each page. (2) Muse Spark 1.3 extracts every requirement with its exact quote, clause number and source page; the quote is verified against the PDF text in Java. (3) A deterministic Java engine (BigDecimal, no floating point) compares each requirement with the company database and returns PASS, FAIL, UNKNOWN (document missing) or NEEDS_REVIEW. (4) It produces a risk register, a missing-document checklist, a rule-based readiness score, a Bid / Conditional Bid / No-Bid recommendation, pre-bid clarification questions, and PDF/Excel reports. (5) A human must approve; the AI never submits a bid.

Key design choice: the AI only extracts and explains; it never decides a verdict. Corrigendum (amendment) handling re-checks only the changed clauses. PDF text is wrapped as untrusted input to resist prompt injection. Every run shows whether the AI components ran LIVE or in FALLBACK mode.

Demo flow: a seeded company vs a 6-page RFP. v1 = NO_BID (score 40.00). After Corrigendum 1 = CONDITIONAL_BID (80.00). After adding the missing solvency certificate = BID (100.00), pending human approval.

Honest status: [N] server tests and [M] client tests passed at the last run. The 70% review-time reduction is a target, not a proven result; measured value: [from docs/SUBMISSION.md, with its assumptions]. Items not completed or not verified are listed in docs/SUBMISSION.md.

## More from Meta Global AI Developer Hackathon

- [Mh0tz](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5276)
- [Walkaway x Muse Blitz](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5277)
- [ZSohaib](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5278)
- [ascension2](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5280)
- [Let Me Solo It](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5281)
- [Phantom](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5282)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5279. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt