NexoProc- Ultimate Procurement Solution.
Primary user: The procurement accountant at a multi-department company, responsible for checking employee purchase requests, comparing supplier quotes, and preparing purchases for approval. What goes wrong today: Requests arrive through email, chat, or spreadsheets, often with missing details. The accountant has to chase employees for specs, check budgets and approval rules in separate places, contact suppliers, compare inconsistent quotes, and follow up with approvers. Routine purchases get delayed, budget commitments are hard to track, and it’s difficult to show why a supplier was chosen.
Preliminary score
- Value for the user18.5 / 25
A specific problem and a clear outcome from solving it.
- Prototype and use of AI21.0 / 30
A working core scenario, and what AI actually contributes to the result.
- Quality testing13.5 / 20
Tests, examples of failures, and comparison with the current approach.
- Feasibility7.0 / 15
Data requirements, running costs and a clear next step.
- Originality4.5 / 10
A meaningful difference from common solutions.
What the judges wrote
- GPT-6 AstraPreliminary68/100
- Strengths
- Your procurement-accountant workflow is specific and covers requests, supplier quotes, approval, purchase orders and delivery. Your documented AI matcher uses supplier-catalog context, explains recommendations and preserves human approval, with an explicitly labelled fallback. Your concrete security regressions and reported fixes—especially competitor-quote isolation and model-invented prices—provide meaningful evidence of engineering discipline.
- To improve
- Show one reproducible request-to-delivery run with actual AI inputs, recommendations and resulting workflow states. Measure accountant time per request and recommendation suitability against both the current manual process and your heuristic fallback, including ambiguous specifications and unavailable products. Add a supplier-data onboarding plan, model and hosting cost estimates, and a scoped customer pilot as your next step. Rotate the included demonstration API key and isolate demo credentials from live company data.
- Claude Opus 5.5Preliminary61/100
- Strengths
- Your problem statement is concrete: a procurement accountant chasing specs, budgets, quotes and approvers across email and spreadsheets is a real, recognisable pain. The prototype covers a full request → AI matching → RFQ → quote selection → auditor approval → PO → delivery workflow, enforced with RLS and server-side RPCs, and localised for Azerbaijani users with AZN defaults. Your testing is unusually honest and specific: the SQL regression suite caught real failures, such as suppliers reading competitors' quotes, cross-company user assignment, weak temp passwords and the AI trusting model-invented prices. The AI step is designed responsibly, with explanations, matching factors, a deterministic fallback and an advisory-only role.
- To improve
- Name and measure the metric the track asks for, such as time from request to PO, number of follow-ups per request, or share of purchases with a documented supplier rationale, and compare it with the current email/spreadsheet process. Evaluate the AI matching itself: build a small labelled set of requests with known correct products and report accuracy, substitution quality and failure cases against the heuristic fallback. Add data and running-cost estimates (DeepSeek tokens per request, Supabase tier) and a concrete pilot next step. Also remove the DeepSeek API key from the committed .env and rotate it, since shipping secrets undermines the strong security work elsewhere.
The team
Oktay Imanzadə Azər oğlu
developer
Orxan MəcidliLead
data