I design and deliver the AI platforms behind some of the world’s largest enterprises.
ArchitectureDeliveryTechnical leadership
Production systems where architecture, delivery, and measurable outcomes had to meet.
View all case studies
A Tredence internal platform using retrieval over an indexed source estate to automate 70% of manual migration work.
Created for Gap Inc. via Tredence in collaboration with Google, GML standardizes how ML and agentic use cases move from development to deployment on GCP.
For Signify Health, a CVS Health company, I built Azure ML foundations, a Snowflake call-center rating application, and an LLM-as-a-judge architecture designed around judge quality per dollar.
An event-driven observability platform on Databricks designed for reusable model onboarding, plus engineering contributions across the client’s customer-facing agentic platform.
Distributed ML and governed data systems across DEKA programs including Roxo, the FedEx SameDay Bot, and the FDA 510(k)-cleared twiist insulin delivery system.
Led migration delivery for AB InBev from Azure to Databricks, landing the target estate with less than 1% deviation from the source.
Production AI is a chain of decisions, and every link is somebody’s job. I work across the whole chain, so quality, cost, governance, and ownership reinforce each other instead of arriving as afterthoughts.
Every strong system starts before the code: turning an ambiguous mandate into architecture, constraints, and a delivery path a team can commit to.
DeliveredBoeing supply-chain AI control tower architecture. ↗I build the reusable path, not the heroic one-off. Shared scaffolding and deployment patterns let the next use case inherit good decisions.
DeliveredGML on GCP for Gap Inc., built with Google and signed off by Google. ↗An unmeasured system is an opinion. I design judge quality and serving cost as one decision, with humans calibrating the judges.
DeliveredCost-aware LLM-as-a-judge evaluation for Signify Health, on Snowflake. ↗Checkpoints, audit trails, and clear escalation live in the architecture, not in a policy document nobody opens.
DeliveredHIPAA-compliant ML delivery at DEKA. Human checkpoints in the migration platform. ↗A system you cannot observe is a system you do not own. Observability and drift detection are part of the build, not the postmortem.
DeliveredTurnkey ML observability for DICK’S Sporting Goods. Onboarding got 10× faster. ↗Prove it once, then make it repeatable. The measure of an architecture is what the second and tenth use cases cost.
DeliveredInternal platform: 70% of migration work automated. AB InBev: under 1% deviation. ↗The best AI systems I have shipped made people faster and sharper, not redundant. That is a design choice, and I make it on purpose.
The migration platform keeps engineers at every checkpoint. The judge escalates to people. The agent hands off. Trust is architecture, not a demo promise.
Evaluation and observability are how a system earns the right to run in production. Numbers hold up in outages, audits, and reviews. Assumptions do not.
One good framework quietly improves every project that follows it. I optimize for the tenth use case, not the flashy first.
Engineers, executives, and compliance partners hear the same system described in the language each of them can act on. That is how AI work earns trust.
I have mentored multiple teams across the ML and GenAI practice, and everything I learn goes into a free public book. Knowledge kept in one head does not scale.
Every stop taught the same lesson from a different angle: the interesting problems live in the whole system, not the model alone.
At DRDO HEMRL, I applied machine learning to solid rocket propellant analysis while studying mechanical engineering. It was my first lesson in systems where the details are unforgiving.
ML and NLP inside live CRM systems, with real-time Kafka and Spark pipelines. Where my models met users for the first time.
An M.S. in Computer Science on a merit scholarship, building multimodal datasets and training video-to-text models before the LLM era. I train models as well as orchestrate them.
Production ML for an autonomous delivery robot built with FedEx, and for an insulin delivery system that is now FDA 510(k)-cleared. Quality was the product.
Agentic platforms, GenAI evaluation, and AI architecture for some of the largest enterprises in the world, with multiple teams mentored along the way.