Inspiration
Small engineering and contracting businesses must reconcile scattered certificates, project records, and qualification documents before a tender deadline. A plausible AI summary can hide a missing requirement or expired document.
What it does
TenderProof turns one authorized tender checklist and a bounded NAS evidence folder into a source-linked requirement matrix: supported, partial, conflicting, not-found, or unreadable. Every requirement stays visible. Findings need returned source quotations and locators when available; contradictions and coverage gaps become a prioritized human-review checklist, not an eligibility verdict.
How it is designed
The user selects the requirement document, evidence folder, and review date, authorizes read scope, reviews extracted requirements, and inspects the matrix and coverage ledger. Saving the exact report locally requires separate approval. A Markdown orchestration Skill delegates ug-filesystem to file-agent, ug-document to document-agent, and ug-knowledge to knowledge-agent. Only real completed expert outputs enter dependent steps. Review scope is capped at 50 evidence documents; source files are never moved, renamed, or deleted. The configured local model is preferred, with no requested external integration or data transfer.
Current status and challenges
This is a Round 1 initial Skill, not a tested NAS application. The Markdown workflow, evidence/status rules, synthetic validation scenarios, and development disclosure are drafted against the official creation-tool catalog. NAS installation, expert runtime behavior, extraction quality, and save verification remain untested and require finalist access. Key design challenges are preserving missing evidence, exposing conflicting dates, and treating OCR failures as incomplete coverage rather than positive findings.
Validation plan
Test complete synthetic evidence for source-supported rows; test a missing certificate plus conflicting validity dates to ensure neither becomes a clean pass; and test unreadable/access-denied files for explicit incomplete coverage. Verify that refused saving creates no output and source files remain unchanged. Results are limited to reviewed scope, not legal compliance, authenticity, or tender acceptance.
Development disclosure
Original workflow newly developed for this Hackathon. Cue AI assisted ideation, drafting, and static review. UGREEN's official UGOS Skill Creation Tool, uliya-skill-kit 0.7.3-en, and capability catalog 2.0.0 are development references under event-provided terms; the kit is not redistributed. No third-party code, weights, libraries, datasets, or external services are bundled. Runtime depends on UGREEN's registered experts and configured model. No device execution is claimed.
