Home/Work
Work

Outcomes,
not slideware.

A few of the production systems we have built — what we shipped, in what timeframe, and what it changed for the business.

Financial services

Real-time risk decisioning at 14ms p99

A top-20 European bank was running fraud and risk decisions on a nightly batch — leaving an 18-hour window of exposure. We replaced the batch with a streaming feature platform and an online inference layer, with full lineage from decision back to source events. The team now ships new risk policies in days instead of quarters.

14msp99 latency
$31MAnnual loss avoidance
9 moKickoff to GA
Insurance

Claims triage with document AI

A multi-line insurer needed to compress claims cycle times without sacrificing audit. We built a document intelligence pipeline plus an operator console for first-notice-of-loss — confidence-scored extraction, deterministic routing, and human review where it mattered. Routine claims now resolve in minutes; complex claims reach the right adjuster on the first hop.

82%Auto-triaged
Faster cycle time
+18 NPSClaimant CSAT
Manufacturing

Predictive maintenance across 1,200 assets

A specialty manufacturer was losing weeks of production a year to unplanned downtime. We built edge telemetry, a lakehouse, and forecasting models — surfacing failures 11 days ahead and integrating directly into the work order system. Maintenance shifted from reactive to scheduled, with full traceability per asset.

11dMedian early warning
−37%Unplanned downtime
1.2kAssets monitored
Retail

Demand forecasting across 4,800 SKUs

A specialty retailer was overstocking long-tail SKUs and stocking out on hits. We built a hierarchical forecasting system tied to the merchandising and replenishment workflows — improving in-stock rates while reducing working capital tied up in slow movers.

+6.2ptIn-stock rate
−14%Inventory days
4.8kSKUs forecast

Have a hard data or AI problem? We'd like to hear it.

Start a project →