ABOUT THE PROJECT

ProcureAI — Project Background

ProcureAI is an AI-powered procurement compliance platform built by BIMSaarthi Technologies for the Government of Andhra Pradesh. The platform spans the full procurement lifecycle — drafting → validating → evaluating → communicating — with end-to-end audit trail, DPDP-compliant Telugu translation, and rule citation chain from every claim down to the underlying ValidationFinding / BidEvaluationFinding kg_node.

BIMSaarthi Technologies

DPIIT-registered startup · Mangalagiri Innovation Hub, Andhra Pradesh

BIMSaarthi Technologies operates from the Mangalagiri Innovation Hub in Andhra Pradesh, focused on building AI tooling for government workflows that respect Indian regulatory regimes (DPDP, CVC vigilance norms, AP-State procurement standards). This platform is our submission to the RTGS Hackathon 2026 — Real-Time Governance through Software, hosted by the Government of Andhra Pradesh.

International Reference Systems

Systems we studied while designing the procurement-AI stack

Czech Republic
ALICE

AI-based contract analysis at the Office for the Protection of Competition. Reference for our rule-citation chain + violation typology pattern.

Brazil
INACIA

Audit assistant deployed at the Federal Court of Accounts. Reference for our composite-finding pattern with sub-aspect breakdown.

Singapore
AIPA

Government Technology Agency procurement AI. Reference for our effective-L1 computation accounting for ALB + cartel-suspect skip chains.

Technology Stack

Knowledge graph
  • →Supabase PostgreSQL
  • →kg_nodes (JSONB additive)
  • →kg_edges + fact_sheets
Validation pipelines
  • →BGE-M3 embeddings
  • →OpenRouter (qwen-2.5-72b)
  • →Three-valued condition_when
  • →L24 evidence guards
Module 4 Communicator
  • →Sarvam-M /translate API
  • →DPDP pseudonymisation
  • →Filesystem cache (SHA256)
  • →EN+TE bilingual output
Frontend
  • →Next.js 14 (App Router)
  • →Tailwind CSS
  • →React 18 server components
  • →Vercel deploy
Reports
  • →python-docx
  • →reportlab PDF (L75)
  • →Markdown intermediary
  • →5-layer drilldown chain

Compliance Posture

The platform is designed for production deployment in AP State government infrastructure. Compliance is built in at every layer:

  • ✓DPDP Act 2023: bidder PII (PAN, GSTIN, mobile, email, address, signatory) pseudonymised before crossing any external API boundary; pairs sorted longest-first to avoid substring collisions; restored after translation
  • ✓CVC vigilance: cartel-suspect detection, ALB corroboration, separation of bidder-facing communications from internal vigilance reasoning
  • ✓AP-State norms: AP-GO-094/2003, AP-GO-062 (ABC M=2), AP-GO-089 (12-month solvency), MPG-255, AP-PROC-* seeded rules
  • ✓Audit defensibility: deterministic audit_id (SHA256 of source kg_nodes) on every communication; 5-layer drilldown chain from ComparativeStatement → BidAnomalyFinding/EligibilityMatrix/TenderRanking → BidEvaluationFinding → BidSubmission/BidderProfile → fact_sheets
  • ✓Bilingual operations: Telugu support for bidder-facing communications via Sarvam-M (India-hosted, in-country data residency); internal communications English-only

Architecture & Lessons Archive

All architectural decisions documented in LESSONS_LEARNED.md (89 entries, L01–L89). Notable patterns:

  • L80: Composite-finding pattern (N sub-aspects evaluated together)
  • L81: JV-aware validator with cross-profile lookup
  • L82: Module 3 Extensions arc completion + DemoBidder pattern
  • L83: PDF renderer integration via reportlab
  • L84: Module 4 Communication architecture design spec
  • L85: M4.2 drafter pilot pattern
  • L86: JSONB merge via fetch-modify-patch (PostgREST limitation)
  • L87: Sarvam-M Telugu integration with DPDP pseudonymisation
  • L88: 6 remaining communication types (pattern stability)
  • L89: Q&A 2-direction workflow with parent_communication_id threading