Tender Radar: finding government tenders for small contractors, entirely on their own laptop
Описание и применение
Tender Radar позволяет малым подрядчикам быстро оценивать государственные тендеры, проверять соответствие требованиям и готовить документы локально, используя Gemma 4 и TabPFN.
Задача
Small electrical contractors need to quickly assess government tender notices for eligibility, compliance, and win probability, but lack tools to do this locally and securely.
Результат
On an M3 laptop with 8 GB RAM, Tender Radar processed 8 tenders in 22 minutes, extracting 51 clauses, producing 24 met, 5 partial, 19 missing verdicts, and drafting a bid pack with 8 unsupported sentences removed. Win‑probability AUC improved from 0.731 (sklearn baseline) to 0.752 (TabPFN).
Рабочий процесс
- Clone the repository
- Generate synthetic demo corpus
- Load corpus into local database
- Scan tender notices
- Open the dashboard to review board and details
- Draft bid pack
Доказательства
Tender Radar answers all three on the contractor's own machine: Extracts the eligibility clauses…
Measured on an M3 laptop with 8 GB RAM, everything local: 51 clauses extracted across 8 tenders… 22 minutes for 8 tenders.
Win model ablation on the same time‑ordered split (147 submitted bids): sklearn baseline AUC 0.731 / Brier 0.260 vs TabPFN v3.5 + thinking: AUC 0.752 / Brier 0.209.
- Дата записи: 2026-10-17T15:56:13.293Z