Nazar Sopiha AI Tech Lead & Builder | Berlin

GenAI in production

2024-2026 / Two systems / Daily production use

The through-line of my GenAI work: systems that reach production and stay there - validation gates, provenance, and daily users instead of demo theater.

Production platform for industrial engineering

A German industrial automation OEM processes dense engineering requirement documents as a core part of its business. The company had no GenAI experience and a concrete problem: the document workflows were manual, slow, and dependent on scarce senior engineers.

What I owned

The full arc of a greenfield engagement with a customer who had never bought GenAI before:

  • Pre-sales and scope shaping - translating unfamiliar-domain discussions into a concrete Phase-1 commitment
  • Team - staffed a 7-person cross-disciplinary team (solution architect, two developers, DevOps, tester, business analyst, PM): 30+ screens, solo technical interviews, two external hires onboarded
  • Architecture and delivery - LLM extraction, OCR, RAG, and vector retrieval on Azure, shipped as v1.0.0 to production
  • The relationship - single technical anchor through scope evolution, from first workshop to daily production use

Outcome

Phase 1 is live and in daily use by the customer's engineering department. The engagement is paused deliberately at KPI measurement - the gate before scale-out to the customer's parent automation group.

Internal GenAI accelerator

Consulting practices accumulate slideware; engineers trust running software. Over 18 months I built an internal GenAI accelerator as a functioning product - named, branded, with a deployed web UI and an end-to-end user flow - rather than another proof-of-concept notebook. It is consumed across the GenAI practice at GlobalLogic for three jobs: presales demos in front of customers, internal upskilling, and a reference implementation that client-project teams start from.

Why it mattered commercially

I positioned the accelerator into automotive presales conversations - using embedded-automotive fluency to align demos with how OEM and Tier-1 development workflows actually look, and presenting it live to external stakeholders. It carried product vision, architecture, and a 4-5 person engineering team's worth of iterations under one roof.