Case studies

Measured by outcomes, not lines of code.

Products and systems I have designed and shipped, from a financial intelligence platform to on-device control and granted IP.

Financial intelligence · Founder

StockLens

A multi-agent pipeline that turns fragmented SEC and policy filings into auditable signal.

Challenge: Turn fragmented, high-volume SEC filings and government policy releases into structured intelligence that survives an audit, without losing signal or slowing down as the corpus grows.
Execution: Designed and built a deterministic, staged multi-agent platform. Independent ingestion agents per source feed a shared data foundation, and every output carries an evidence trail back to the original filing. Heavy processing runs on owned hardware behind a thin cloud serving layer.
Outcome: Scaled past 15 million structured artifacts across roughly 970,000 SEC entities, and launched commercially with live subscription billing.

970kSEC entities

15M+artifacts

85-90%extraction accuracy

sub-secondretrieval

$49/molive Stripe billing

Edge AI · Data-centre cooling · IoTX

An AI-native control loop for data-centre cooling that runs entirely on the device.

Challenge: Data-centre cooling has to react to fast thermal and electrical swings in dense compute fleets, under fixed power budgets, with no tolerance for a missed control cycle and no dependency on the cloud.
Execution: Architected an on-device platform that runs the full closed control loop locally. Safety and system-health functions are always on by design, and the system degrades gracefully long before it can miss its timing budget.
Outcome: Worst-case execution under 1 ms against a 2.0 ms limit on an ARM Cortex-M33 target, with up to 32% compute saved per cycle. Patent filed, and in pilot with EU data-centre operators under NDA.

sub-1 msWCET vs 2.0 ms limit

up to 32%compute saved per cycle

Cortex-M33ARM on-device target

Client details under NDA.

Computer vision · Granted IP

A camera system that finds the stroboscopic effect on its own. Two granted US patents.

Challenge: Stroboscopic inspection lets engineers see vibration in fast-moving machinery, but locking onto the right strobe point has depended on a trained operator.
Execution: Co-invented a camera-based system that detects automatically when the stroboscopic effect is reached, then analyses the frozen frames to identify the object and measure its movement or vibration.
Outcome: Two granted US patents, co-invented with Abraham Greenboim and assigned to IoT Technologies LLC.

2granted US patents

B1 + B2USPTO grants

Next step

Technical due diligence, fractional CTO work and edge AI builds.