BidBridge
BidBridge is built for bid managers, business development leads, and SME owners at companies that regularly compete for public-sector and enterprise tenders, starting with suppliers in Azerbaijan and later the wider Caucasus and MENA region. Today, these teams manually read lengthy tender packages, extract mandatory requirements, and cross-check them against scattered company records such as certificates, past projects, financials, and staff credentials. This makes the initial bid/no-bid decision slow and error-prone: teams can waste hours pursuing tenders they are not qualified for, or miss a single mandatory requirement that puts the entire bid at risk of rejection.
Preliminary score
- Value for the user20.5 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI23.0 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing16.5 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility11.5 / 15
Data requirements, running costs and a clear next step.
- Originality6.5 / 10
A meaningful difference from common solutions.
A finalist: the live jury scored it 72.0 / 100 on stage.
What the judges wrote
- GPT-6 AstraPreliminary78/100
- Strengths
- Your focus on first-pass bid/no-bid screening addresses a concrete, costly workflow for tender-active SMEs. Your rules-first checks, source-quote verification and explicit UNKNOWN state make AI's contribution auditable rather than merely conversational. Your reported testing is unusually specific: the 15 logged failures and caught regression show meaningful investigation of reliability.
- To improve
- Your next evaluation should use procurement-expert-reviewed labels across more tenders, including Azerbaijani, and compare decision accuracy and review time against manual screening. Your README still lists the dashboard, evaluation harness and end-to-end test as unfinished; reconcile this with the submission and provide a reproducible, recorded full workflow. Your feasibility plan should validate paid-tier operating costs and explain how confidential supplier records will be protected during a small customer pilot.
- Claude Opus 5.5Preliminary78/100
- Strengths
- Your evaluation is unusually honest and specific. You report citation validity, recall ranges across runs, and compliance accuracy on two real tenders, and you log 15 concrete failures, such as the EUR vs AZN comparison and the regression caused by your own prompt fix. Your architecture splits the work sensibly: deterministic rules, scoring and vetoes live in code, and the LLM handles the language. You add quote verification and no unsupported positives, and the system prefers UNKNOWN over guessing. You also cut API calls from about 28 to 6–10 per analysis and priced a run at about 3 cents.
- To improve
- You have not yet measured how long a bid manager takes to do the bid/no-bid check by hand, so you have no adoption metric. Timing even 2–3 real bid managers on the same tenders would turn your claim into evidence. Your ground-truth labels were drafted by AI and cover only two English tenders, so have a human expert review them and add Azerbaijani tenders, which is your stated beachhead. Your README status section still says the frontend and evaluation harness are unbuilt, and it names Anthropic while the submission says Gemini. Update it so reviewers can trust that the repository matches the demo.
The team
Ruslan Sadigov
developer
Emil Ahmadov
data
Eldar HasanliLead
product