From AI engineer to infrastructure engineer — and back.
I've spent years learning how systems actually work at scale. OmniTechnicus is where I write the lessons down.
Twenty years ago, I stood at the intersection of curiosity and innovation, doing PhD research on neural networks at University POLITEHNICA of Bucharest — captivated by one question: how does data shape the way machines learn and behave?
While my peers focused on theoretical models, I was poking at patterns we’d now call data poisoning and prompt engineering. I didn’t know it then, but I was asking questions that wouldn’t go mainstream for two decades.
The pivot that changed everything
In 2007, Microsoft knocked. I faced a choice: keep pushing toward a PhD in a field the world wasn’t ready for, or apply AI/ML thinking to real production systems. I took my Master’s and dove into platform engineering. What looked like a departure from AI was actually the start of a deeper journey — learning how distributed systems behave at scale, how to design resilient infrastructure, how to lead teams through genuinely hard problems.
Building the foundation
At Microsoft I started in Windows Phone platform engineering — designing test frameworks, then leading device updates and the Windows Insider app — and learned what it takes to ship software that millions of people depend on. I discovered that great engineering isn’t just writing code: it’s understanding systems, anticipating failures, and building for scale.
Then I moved into virtualization and cloud infrastructure: Hyper-V performance, validation frameworks for Azure services like Event Grid, Key Vault, and Cosmos DB, CI/CD automation that cut deployments from two and a half hours to thirty minutes, and systems that recovered from regional outages in thirty minutes instead of days.
Each project taught me something new about resilience, automation, and the balance between innovation and reliability.
Great engineering is not just writing code. It is understanding systems, anticipating failure, and building for scale.
Full circle: AI returns
Then came 2020, and AI exploded into the mainstream. Suddenly the questions from my PhD research were everywhere — and I had something rare to bring: thirteen years of production engineering to apply to them.
I built a production RAG system that tripled documentation-match accuracy to 77% on 20–30 live incidents a day. I architected an Artifact Cache Monitor that reduced incidents by 90% and bandwidth waste by 99.8%, gave an intern ownership of predictive worker scaling that shipped with 25–30% resource savings, and architected the persistence layer for the Microsoft Copilot orchestrator. The researcher who left academia in 2005 came back — not with theories, but with production-grade AI systems running at Microsoft scale.
What drives me today
In July 2026 I left Microsoft to work on AI systems full time.
I don’t just build AI systems; I bridge worlds — connecting cutting-edge research with the messy reality of production. I take ideas out of papers and turn them into things that survive real traffic, real failures, real constraints. My edge isn’t just understanding neural networks or Kubernetes or distributed systems — it’s understanding how they fit together, how to make them work in production, and how to lead teams through the complexity.
I hold a patent for objective application-rating systems and submitted another for AI-driven predictive scaling. But the patents aren’t the point. The point is solving problems that matter and mentoring the people who’ll take this further than I will.
The thread
The questions I asked in 2005 are still the questions I ask today. The difference is that now I have the tools to answer them — and a place to write the answers down honestly.
Career
-
Independent AI Engineer
OmniTechnicus
Left Microsoft in July 2026 to work on AI systems full time — building them, testing what holds up, and writing the lessons down here.
4 highlights
- Writes OmniTechnicus: long-form articles on agentic architecture, quantization and AI governance, with corrections made in public
- Built and runs OASIS, the emergency operations platform for the City of Issaquah, WA: offline ICS forms, activity monitoring and Winlink radio messaging, designed to work with no internet at all
- Leads the city’s volunteer radio communications team; licensed amateur radio operator W4MHI
- Maintains weather-chart-card-ha, a Home Assistant card with more than 2,000 installs through HACS
-
Principal Software Engineer / Architect
Microsoft
AI and LLM systems on a platform serving 200+ microservices, ending with the orchestrator behind Microsoft Copilot.
7 highlights
- Architected and shipped the persistence layer for the Copilot orchestrator on Microsoft Orleans: specialized grains with reminders balance state across production clusters, with zero state loss across deployments and 99.999% recovery through database and cold bootstrap
- Built guardrails into that persistence layer that detect and block sensitive content before it is stored or propagated
- Built a RAG system for incident triage over 20–30 live incidents a day that raised documentation-match accuracy from 25% to 77%, attaching confidence-scored troubleshooting guides automatically
- Designed a parent-child deduplication algorithm that collapsed hundreds of monthly incidents into tens, so a single root-cause fix closes a whole chain
- Designed an ML model that reads branch naming conventions and proposes three ranked configurations, cutting service onboarding from hours to minutes across 200+ microservices
- Hired 10+ engineers and mentored 20–25 over five years: three promoted to Senior, three to SDET2, two recommended for Principal
- Gave an intern ownership of predictive worker scaling; it shipped with 25–30% resource savings, went to patent submission, and the intern joined full time
-
Senior Software Engineer
Microsoft
Led device updates and Windows Insider, owned Hyper-V performance, then built the platform and delivery systems the later AI work ran on.
6 highlights
- 2014–2015: led Windows Phone device updates (scheduled updates, zero rating, the update service, the daily update framework) and the Windows Insider app for Windows 10 preview users, an Azure service with a SQL back end
- 2015–2017: owned Hyper-V boot, network, CPU and I/O performance metrics and the framework behind them, including nested-VM monitoring for Windows and Linux guests and a C/C++ host-to-VM socket tool
- 2017–2021: owned and architected the configuration solution for the Windows update pipeline, and automated package validation for Azure Event Grid, Key Vault, Storage Blob and Cosmos DB
- From 2017, continuing as Principal: led the Artifact Cache Monitor with four engineers for four years, taking incidents from 60 to 6 a month and transfer from 288 GB to 540 MB a day, and designed the APIM failover that recovered from an Azure regional outage in 30 minutes instead of days
- From 2017: built CI/CD for an 80-engineer organization, taking deployments from 2.5 hours to 30 minutes and freeing about 2,600 engineer-hours a year, plus a smoke-test suite generated from production incidents that took deployment defects from 30 a year to 4
- Led and mentored the device update team, including promotions, and a vendor team that validated the daily update process
-
Software Development Engineer
Microsoft
Windows Phone and mobile device management, after a year as a contractor in 2006.
4 highlights
- System Center Mobile Device Manager (2007–2008): test frameworks and device-management tooling for Microsoft’s enterprise mobile platform
- Windows Phone backup, migration and restore (2008–2012): consolidated the Windows Phone 7 and Zune code end to end, added fuzz, stress and reliability suites for the data store, and built a native COM automation framework
- Designed test frameworks for Windows Phone 7 and 7.5 and Zune 4.8, and planned and designed a Windows Phone management tool
- Core OS (2012–2014): built test frameworks for device imaging, flashing and the Zune backup/restore protocol, and a COM server simulator for cloud services hosted on the device; shipped Windows Phone 8
-
Doctoral Researcher – Neural Networks & Human-AI Interaction
University POLITEHNICA of Bucharest
Early research on data influence in neural networks, preceding modern understanding of data poisoning and prompt engineering
4 highlights
- Pioneered research on how data patterns influence neural network learning and behavior
- Explored human-AI interaction patterns in neural network systems
- Left the doctoral program in 2005, before completing it, for a Master of Science in Computer Science at Maharishi International University (2005–2008)
- Bachelor of Science in Computer Electronics, University POLITEHNICA of Bucharest (1991–1996); Machine Learning Specialization, Stanford Online and DeepLearning.AI (2025)
-
Patents & Recognition
Microsoft
5 highlights
- US Patent: "Objective Application Rating" (filed 2013, granted 2016) – an automated system that computes objective application quality ratings from telemetry collected across distributed devices
- Patent submission (2024): AI-driven predictive worker scaling for distributed systems – a machine-learning system that anticipates workload patterns and provisions compute dynamically, at 25–30% resource savings
- Microsoft Excellence Award (2021): an equity award for exceptional contributions to platform infrastructure and engineering excellence
- Microsoft Excellence Award (2014): exceptional contributions to Windows Phone
- Goldstar Award (2008): early-career recognition for Mobile Device Management contributions
