Home / Our Work / 12 legacy systems. Month-end close taking 5 days. Zero audit trail.
Manual ID Checks Exposed an Insurer to Fraud: A Vision Agent Verifying in Seconds
CLEINT :
Neutrinos: Insurance Provider
Location :
India
100%
AI-driven checks: subjective manual review removed
3 min → <10 sec
Per document, over 18x faster than the manual check
93%+
Thumbprint detection accuracy on scanned documents
Rural-Ready
Consistent checks across high-volume agent networks
The Hidden Problem
Thumbprint verification on scanned policy documents was done by hand. It looked like a routine step. It was slow, inconsistent and open to both fraud and human error, at three minutes per document. Scan quality varied wildly, and onboarding ran through high-volume, rural and offline-first agent networks. A human reviewer cannot hold a consistent standard at that scale, so every added document increased the load and the room for a missed mismatch. That exposed the insurer to fraudulent onboarding and to policies issued on checks no one could defend, in exactly the channels where volume was growing.
What we built
Neutrinos, a computer-vision verification agent that runs the check end to end. A preprocessing stage on OpenCV cleans noisy, skewed and low-quality scans. A detection model locates the thumbprint inside complex document layouts. A CNN on TensorFlow performs minutiae-based matching and returns a confidence score. Above threshold, the agent verifies autonomously. Below it, the case is routed to a human with the evidence attached, so people decide only the uncertain cases. Results surface through a secure upload portal and a real-time dashboard, for single-policy checks and batch runs alike. The subjective manual check is gone, detection accuracy is above 93% on scanned documents, and verification dropped from 3 minutes to under 10 seconds per document, ready for rural onboarding at volume.
"Verification used to be our biggest bottleneck — three minutes per document, with room for human error at every step. Neutrinos gave us seconds, not minutes, and a level of accuracy we couldn't get manually, even in our most rural onboarding channels."
Neutrinos Team
Explore the service behind this work.
AI Transformation
Multi-agent systems, RAG, evals and LLM applications. We build the agents that read, decide, act and verify inside your operations, and the governance that lets you trust them in production.
Get started
Working on something similar?
Let’s discuss your processes and identify practical opportunities to improve efficiency.
30 minutes · No obligation · Senior consultant