ARCHIVES

Indian police stations still rely on paper registers and unencrypted pen drives to store case evidence, a practice that frequently leads to acquittals when courts reject files whose integrity cannot be proven. We built CrimeIntellX to close this gap. At its core is a cryptographic evidence vault that encrypts every uploaded file with AES-256-CBC, fingerprints it with BLAKE3, and anchors the hash inside a Merkle tree so that even a single byte change is caught instantly. The same module registers criminals with ArcFace-based facial embeddings and matches them in under a millisecond through FAISS. A second component, Threat-Intel- Live, pulls ransomware leak posts from RansomWatch, searches Tor via Ahmia, and cross-checks IPs and URLs against AbuseIPDB and VirusTotal. The third component is a natural-language query engine backed by Llama 3.3 70B and a Pinecone vector store; when an officertypes a case description, the engine returns similar FIRs, matching criminals, CDR anomalies, dark-web hits, and an evidence-gap checklist, all within roughly eight seconds. We developed the backend in FastAPI (83 endpoints) and the frontend in Flutter (18 screens), tested with simulated Indian datasets, and verified compliance with BSA Section 65B.