If it can't be measured, it's not real. If it can't be reproduced, it's broken. I build systems that stay correct when the traffic triples.
2002→2026
18+ yrs C# & production engineering
01. About
I design boring, composable primitives that survive production.
15+ years in production engineering. I don't chase hype cycles — I build systems that stay correct when the traffic triples, the regime shifts, or the intern pushes to main. Engagements concentrate around systems that have outgrown their architecture, optimization layers rebuilt natively in .NET, and quant infrastructure that has to match live reality — to the tick.
Toptal · Verified Expert in Engineering · Top 3%02. Experience
Nine brands across fragrance, fashion, jewelry, apparel, and B2B — including Pernod Ricard. Event-driven storefronts, state-machine carts, multi-market inventory automation, A/B tests that doubled conversion rates, and an ML-based sizing engine that cut returns.
Industrial data engineering over 200M+ rows and 6 power-plant types on Snowflake. Cursor-based procedures that timed out rebuilt as deterministic pipelines that finish in hours; grid bids became plannable 1–2 years out.
Platform tracking 5M+ trucks: visual workflow builder, geofencing, one database per tenant, and query paths taken from seconds to 30ms. GPS route replay with physics-based outlier correction.
Real-time bidding aggregation over billions of RTB records — a queue and worker-pool architecture with throughput faster than the ingest rate.
Seven years building a managed-print-services platform and a shop-management platform end to end. C# / .NET and SQL Server.
Hired on scholarship. ERP for major European automotive brands; queue architecture importing millions of parts. VB6 and SQL Server, where the discipline started.
03. Skills
04. How I Think
Every claim — speed, correctness, reliability — needs a number, a baseline, and a method. "Feels fast" doesn't clear the bar.
No black boxes, no magic pipelines, no AI output without gates and traceability. Git-backed audit trails over opaque orchestration.
Same seed, same result — on your machine, in CI, in production. "Works on my machine" is a design failure, not a quirk.
Hard if/else in an optimization problem is a gradient killer. If the optimizer can't see the slope, you'll never find the minimum.
05. Education
Polytechnic University of Craiova — five-year degree
National College Carol I, Craiova
06. Current Focus
Native C# hyperparameter optimization — open source, MIT
Deterministic fills, crash-aware regimes — paper = live, to the tick
Gated, auditable pipelines — no output without traceability
07. Contact