Nazar Sopiha AI Tech Lead & Builder | Berlin

Embedded & low-level

2020-2024 / ASIL C/D / AUTOSAR Classic

Before LLMs, I spent three years where software failure has physical consequences - automotive ECUs under AUTOSAR Classic, MISRA C:2012, and ISO 26262.

Battery management for an electric pickup

On a Tier-1 supplier's battery management system for an American OEM's electric-pickup program, I owned the thermal-management module: cell monitoring, CAN-based diagnostics, and active thermal control. I grew from IC to module owner over 17 months, wrote ASIL C/D safety-critical firmware, served on the project's Change Control Board, and participated in Automotive SPICE Level 1 and Level 3 assessments. The program reached Start of Production.

BSW integration and a hardware-level bug

On an earlier engagement for a global automotive supplier, I integrated the AUTOSAR basic-software layer on a Renesas RH850 ECU - and root-caused a long-running clock-drift bug between the ECU and a peripheral: a slight crystal-oscillator de-synchronization producing cumulative timing mismatch. Weeks of cross-component evidence gathering across the hardware/firmware boundary, closed in direct coordination with the peripheral-side team.

Greenfield: ambient lighting for a premium sports car

On a 5-person greenfield team building an interior ambient-lighting ECU module for a European premium OEM's high-performance sports-car program, I carried module-owner-style scope and established the team's Automotive SPICE process baseline, with the first formal assessment passed at the 6-month mark.

Why this matters for AI work

Safety-critical process discipline transfers: KPI-driven verdicts, change control, and traceability are exactly what production AI systems lack by default - and what customers buying them actually need.