Start from the system, not the model
I want to understand the workflow, the constraints, and the cost of getting it wrong before talking about the stack.
Applied AI architect, operator, and founder
I build AI systems for hard environments.
My work sits where data is messy, decisions matter, and deployment cannot be an afterthought. I care about systems that have to fit real operations, not look impressive for five minutes.
Over the last decade, I have worked across recommendation systems, speech and NLP, computer vision, generative AI, MLOps, and AI infrastructure. The common thread is simple: build something useful, make it hold up in practice, and be honest about what it takes to get there.
I care about systems that can survive contact with users, data, deadlines, and deployment.
IEEE PES GM
CIGRE Paris
DistribuTECH
AEIC Awards
Charles Steinmetz Award
NVIDIA GTC Blog
How I fit
I do my best work where the operating problem is real.
I am not interested in AI theater. The problems that hold my attention are the ones that have to clear real constraints: legacy systems, fragmented data, user skepticism, security reviews, cost, and deployment reality.
That is why the work kept moving toward harder environments. In those settings, you do not get credit for the demo. You get credit when the system works, the team trusts it, and the result holds up after rollout.
Working style
The pattern behind the work is consistent.
Own the work end to end
I like carrying things from vague requirements through architecture, delivery, and cleanup instead of handing off half-finished thinking.
Keep trust in the loop
Explainability, safety, workflow fit, and human judgment are part of the build. They are not polish added at the end.
Selected work
Flagship systems across the career arc
Early commercial ML
Recommendation and pricing systems
Built recommendation and pricing systems tied directly to customer behavior, monitoring, and retraining, including work that contributed to a 10% revenue lift.
Global ML deployment
15+ models across multiple regions
Led rollout of recommendation, churn, and forecasting systems across North America, Europe, and Asia, affecting roughly 20 million customers with stronger deployment discipline behind them.
Speech & NLP
Production speech-to-text systems
Built transformer-based transcription systems for live operational use, reaching 98% accuracy while reducing compute cost by 80%.
Mission-critical analytics
Spatiotemporal outage analytics
Built outage analytics that improved location accuracy by 78% and cut detection time from hours to minutes in utility operations.
Computer vision
AI inspection with synthetic data
Built inspection systems that reached 92% defect-detection accuracy, reduced manual inspection time by 70%, and improved field safety.
Generative AI & infrastructure
Knowledge systems and compute optimization
Built retrieval and generative AI systems for regulated teams and optimized HPC infrastructure enough to cut compute cost by up to 90%.
Career arc
The full story explains the current chapter.
I did not start in mission-critical AI. The path moved from early business-facing data systems into MLOps, production ML, speech and NLP, public technical proof, and then into harder enterprise environments where reliability and adoption mattered more.
See the full career arcCommercial systemsRecommendation, pricing, monitoring, revenue-linked ML
Deployment disciplineMLOps, CI/CD, lifecycle standards, global model rollout
Hard environmentsOutage analytics, inspection, retrieval, HPC, regulated operations
Founder chapterSAVYMINDS as the platform expression of that work
Public proof
The recognition matters because it came out of shipped work.
First-author publications, AEIC awards, the Charles Steinmetz Top Innovator Award, and visibility through the NVIDIA GTC Blog and industry outlets all help validate the work. None of that matters without the operational systems underneath it.
See research and recognitionSelected proof
- IEEE PES GM, CIGRE Paris, and DistribuTECH first-author publications
- AEIC Achievement Awards in 2023 and 2024
- Charles Steinmetz Top Innovator Award
- NVIDIA GTC Blog and broader utility-industry visibility
Current chapter
I’m now building SAVYMINDS.
SAVYMINDS starts with a real platform, not a collection of demos. It already has the core ingredients needed to support multiple products responsibly: tenant-aware operational state, connected data, model governance, evaluation, policy, and runtime lanes for different kinds of work.
The first products are deliberately specific, but they are not the whole story. They prove the platform in environments where review, analytics, workflow handoff, and enterprise fit matter from day one.
Learn about SAVYMINDSWhat is true today
Cloud-first. Founder-led. Grounded.
- Shared platform with connected data, runtime control, evaluation, and governance
- Early products focus on screening and customer interaction workflows
- Cloud first, with connected and private deployment paths when the environment requires it
Contact
If you are building something real and want to talk through systems, product, or deployment, reach out.
I’m most interested in conversations around applied AI, hard operational problems, platform direction, and work that has to hold up outside the lab.