CRPF Procurement Prototype

High-Assurance AI for
Formal Audits.

Aegis eliminates manual evaluation bottlenecks. Using deterministic Dual-Pass Normalization, we extract, evaluate, and visually ground vendor evidence against rigid statutory criteria.

Initialize Engine

System Architecture

  • Zero Silent Disqualifications

    The AI never makes final rejection decisions. All ambiguous edge cases, missing data, or proximity flags automatically default to a mandatory Human-in-the-Loop review.

  • Visual Grounding

    Extracted data is deterministically mapped to exact `[x, y]` bounding boxes. Officers click a flagged criterion and the UI jumps to the exact highlighted paragraph in the original PDF.

Dual-Pass Normalization

LLMs extract semantic context. A deterministic Python engine handles mathematical validation. Hallucinations are mathematically impossible.

Proximity Gap Detection

If conflicting financial figures (e.g., Standalone vs. Group Turnover) exist within a 30-word radius, Aegis halts and alerts the officer.

Append-Only Ledger

PostgreSQL constraints forbid updates or deletions. Every human override and algorithmic verdict is logged into a tamper-proof chain.