Captured from the live deployment, not a mockup
01 · The problem
The Karnataka State Crime Records Bureau stated that fragmented station-level data blocked state-wide analysis. Scoped one product where a station sees its own slice and the bureau reads the whole state off the same graph.
02 · How it works
- 01
Shipped end to end over 50 days on Zoho Catalyst: command dashboard, case-linkage graph, offender watchlist, drillable map and a bilingual assistant, reaching Level 2 of the datathon.
- 02
Chose to surface shared modus operandi as a ranked hypothesis rather than a name, so the tool could not read as an accusation, and paired every flagged case with a reason and a next action.
- 03
Built for users who cannot debug a model: a station officer gets a ranked case queue with a plain-language reason, while the bureau reads state-wide analysis off the same graph of 59,985 FIRs.
- 04
Set the responsible-AI boundary explicitly, ranking shared modus operandi as a hypothesis with a glass-box factor breakdown rather than naming a suspect.
03 · What it cost, and what it returned
Shipped the full loop on Zoho Catalyst in 50 days, including an 0.870 AUC forecast, DBSCAN hotspots and evidence OCR, reaching Level 2 of the KSP Datathon.