Kubera
Small suppliers often have to review lengthy tender documents and search through separate company files to find evidence for each requirement. This manual process is time-consuming and makes it easy to miss missing, expired, or unclear documents. Kubera helps suppliers review requirements and match them with supporting company documents.
Preliminary score
- Value for the user18.0 / 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 testing9.0 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility10.0 / 15
Data requirements, running costs and a clear next step.
- Originality6.5 / 10
A meaningful difference from common solutions.
What the judges wrote
- GPT-6 AstraPreliminary67/100
- Strengths
- Your workflow addresses a concrete burden for small suppliers, connecting tender requirements to supporting evidence and a human-reviewed CSV checklist. Your documented page-level quote verification, unverified-source labels, and distinction between missing evidence and missing company documents are valuable safeguards. Your multilingual interface and synthetic sample pack make the prototype relevant and safer to try.
- To improve
- Your testing would be stronger with completed sample outputs and expected-versus-actual results against your versioned ground truth, including missed requirements and incorrect evidence matches. Your claim that scanned-PDF issues are fixed needs a reproducible example and clarification of how scans are handled when OCR is disabled by default. Your adoption case needs a timed comparison against manual review and a measurable target for time saved without increasing omissions. Your production plan should quantify model and hosting costs per analysis and prioritize authenticated, isolated workspaces before accepting confidential documents.
- Claude Opus 5.5Preliminary66/100
- Strengths
- Your target user is specific: small suppliers in Azerbaijan matching tender requirements to their own company files. The pain of missing, expired or unclear documents is concrete. Your README describes a thoughtfully engineered pipeline with structured Gemini output and server-side re-validation. Each quote is checked against the extracted page text, and unverifiable findings are labelled "Source not verified". Human review states, CSV export, a trilingual interface, a synthetic sample pack with versioned ground truth, and a deployed Railway demo all suggest a real end-to-end workflow rather than a chatbot wrapper.
- To improve
- Your quality testing write-up is thin. Three sample tenders and "we solved it!" give no numbers, so report requirement-extraction recall and precision against your ground-truth fixtures, the share of quotes that pass verification, and concrete failure cases. Your submission says scanned PDFs were fixed, but the README says OCR is off by default and scanned pages stay unread, so clarify which is true. The track asks for an adoption metric, so measure time per tender against manual review, and state the per-analysis Gemini cost alongside your 19 AZN demo price.
The team
Vaqif Məmmədli
product
Tural Nasirov
data
Elnur MəmmədovLead
developer
Safar Gahramanli
developer
Nihad İsmayılov
business