Automotive AI

Automotive is where my two backgrounds meet: years of safety-critical ECU firmware (the Embedded & low-level direction) and applied AI work. Today that combination lands in one place: evaluating the AI compute stack behind driver assistance.

GPU evaluation for automotive perception

A Tier-1 automotive supplier needed a defensible answer to an expensive question: which data-center GPU should run the AI inference stack behind their next-generation driver-assistance perception - AMD Instinct MI300X or NVIDIA H100 NVL?

What I owned

Everything between "vague pre-sales conversation" and "signed-off final report":

• Pre-sales and scoping - turning an ambiguous hardware question into an engagement with defined KPIs and deliverables
• Team - drafted the role specs, screened 30+ candidates, ran the technical interviews solo, and onboarded a 6-person team of external hires and internal transfers
• Technical direction - the benchmark design across 2D detection, 3D surround-view fusion, and CUDA-to-ROCm portability
• The customer relationship - steering through requirements churn as the single technical owner, from kickoff to the 76-page final report

Outcome

Six months, six people, one clear verdict. The report feeds directly into the customer's next hardware evaluation cycle for its perception stack - the kind of decision that outlives the engagement by years.