India's cyber-fraud judgments grew 533% in five years.

Ingest NCRP leads, bank s.94 BNSS notice responses, and mule account flags from every state cyber cell. Build a unified graph of phone numbers, UPI VPAs, bank accounts, IMEIs, and wallet addresses — with community detection revealing rings that span 5+ states. Today, each cyber cell sees only its slice. This layer lets any investigator see the full footprint the moment they flag one node.

A Sarvam / Llama-Indic model deployed inside state police infrastructure. Semantic search over FIRs and chargesheets in Hindi and regional languages. Auto-drafts BNSS s.94 production orders and s.106 freeze requests. Summarizes phone dumps and WhatsApp exports. Never sends case data to external APIs — runs entirely on-prem.

Visual money-trail reconstruction across UPI rails, bank transfers, wallet layering, and crypto off-ramps. Plug in bank notice responses under BNSS, overlay on-chain data from public explorers. Turn a weeks-long manual reconstruction into a 30-minute visual.

Continuously monitor public Telegram scam-recruitment channels, fraud-VPA dumps, Play Store 1-star reviews, and news feeds. Auto-flag new mule VPAs and phishing domains in real-time. Cross-reference with live NCRP leads to identify networks while they are still active, not after the money is gone.

Given the 2.3M annual NCRP volume, no state has the staff to investigate every complaint. A triage layer scores incoming complaints by recoverability — how fresh the trail is, whether the mule is still active, whether a bank notice is likely to succeed. Officers work the highest-return cases first.

Feed judgment outcomes from Indian Kanoon (what this demo already uses) back into the investigation pipeline. Which evidence types lead to conviction? Which courts move fastest? What section pairings work best? This turns every judgment into institutional knowledge instead of leaving it buried in PDFs.