AI Engineer · Computer Vision · Open to work
Making machines see, learn,
and scale.
Eight years of production AI. Training detection models on satellite imagery, shipping LLM curricula to graduate classrooms, and stitching IoT automation across industrial plants that long predate the word agentic.
The short version.
I build AI systems that spend more time in production than in demos. My work lives at the intersection of computer vision, deep learning, and the less glamorous engineering that keeps either one actually useful — FastAPI services, Docker pipelines, plant-floor PLCs, HPC job schedulers. The boring parts. The parts that don't show up in a Medium article.
Currently I coordinate the Data Science & LLM curriculum at the University of Pittsburgh, where I mentor master's students on RAG architectures, fine-tuning, and the art of debugging a model that should be working. Before Pitt, years in the field: automating fifteen manufacturing facilities, building predictive-maintenance pipelines for HVAC equipment, and training satellite-imagery detectors that cut a pilot client's manual survey cost by ninety percent.
I hold an M.S. in Information Science (AI specialization) from the University of Pittsburgh. My French Talent Passport Famille visa means I'm work-authorized from day one in Paris. I'm pragmatic, measured, and mildly allergic to ML pitches that won't survive contact with real data.
If you're hiring for a role where the model has to actually ship — on real hardware, to real users, under real budgets — we should probably talk.
Six projects that shipped.
RTU Detection Platform
Full-stack rooftop HVAC detection over satellite imagery for Airoverse. A YOLOv11 + RF-DETR ensemble reached 90% precision, 80% recall; the deployed pipeline cut the pilot client's manual survey cost by ninety percent. Geospatial via Google Maps, Roboflow for data curation, FastAPI behind a React console, Docker to GCP with CI/CD.
FastAPI · React
Docker · GCP
RetailVision Analytics
Real-time in-store analytics pipeline. Customer tracking, dwell-time heat maps, and behavioural aggregation powered by YOLOv8 + ByteTrack on edge hardware. Built to answer the question every retailer eventually asks: where do people actually go?
Python · OpenCV
Edge deployment
LLM API Gateway
A high-performance gateway in Go for LLM traffic. Intelligent rate limiting, per-model load balancing, fallback routing across OpenAI, Anthropic, and local vLLM endpoints. Written because the off-the-shelf options were all opinionated in the wrong direction.
OpenAPI · Docker
Multi-model routing
Deepfake Detection
A cyber-forensics classifier using InceptionResNetV2 with aggressive augmentation. Real-time TensorFlow inference, tuned to surface the low-confidence boundary cases where human review actually adds value. Graduate coursework; deployed as a class tool.
InceptionResNetV2
OpenCV
Facial Recognition ATM
A biometric authentication prototype using a Siamese network for face verification. PostgreSQL-backed identity store, FastAPI service, full Docker deployment. A working sketch of what happens when you take the card out of card-based authentication.
FastAPI · Postgres
Docker
Stock Bull Trading Platform
Full-stack virtual trading app — live WebSocket market updates, portfolio management, and a Monte Carlo option pricer written from scratch. A long weekend project that turned into a useful teaching tool.
MongoDB · Monte Carlo
Ten years, five chapters.
— Present
Course Coordinator — Data Science & LLM
University of Pittsburgh Pittsburgh, PA
Mentoring 30+ master's students on Generative AI, RAG, and LLM fine-tuning. Previously facilitated applied data-science courses for 300+ students across Python, Tableau, and Power BI. Supervised ten project teams applying scikit-learn and TensorFlow on real datasets.
— Apr 2025
Software Developer — AI / Computer Vision
Airoverse Pittsburgh, PA
Built an AI-powered HVAC rooftop unit detection platform end to end. Trained YOLOv8 and RF-DETR models achieving 90% precision on satellite imagery; integrated Google Maps and Roboflow APIs; shipped a FastAPI + React stack to GCP. Reduced manual detection cost for the pilot client by 90%.
— Dec 2022
Data Automation Engineer
Shraddha Engineering Coimbatore, India
Led IoT and automation initiatives across fifteen facilities, saving an estimated 2,000+ hours annually through ETL and pipeline efficiency work. Built real-time monitoring tools that reduced downtime by 30%. Automated reporting workflows, cutting processing times in half.
— Jun 2020
Systems Analytics Engineer
Blue Star Limited Coimbatore, India
ML-driven predictive maintenance, reducing failures by 22% across commercial HVAC deployments. Built out PLC/SCADA automation across 25+ projects (18% efficiency gain). Led a cross-functional team of ten on analytics and sensor-network deployments.
— Apr 2018
Automation Engineer
Chennai Engineering Services Coimbatore, India
Automated data acquisition and reporting using Python and SQL, improving reliability by 15%. Designed control logic for commercial electrical projects and shaved 10% off project delivery times. First real exposure to the gap between what's elegant on paper and what survives the field.
What I reach for.
PyTorch first for new work; TensorFlow where the team's already there. Fluent with the modern detection stack — YOLOv8, YOLOv11, RF-DETR, Roboflow — and the classic scikit-learn toolkit for everything before neural networks were the obvious answer.
RAG pipelines, fine-tuning, agentic workflows. LangChain and HuggingFace for composition; MLX when the work's local; vLLM and TGI when the work's serious. Comfortable designing around context windows instead of fighting them.
Python daily; Go for services where latency matters; Java and JavaScript/TypeScript when the job calls for them. SQL fluently, including the ugly parts.
FastAPI as my default; Node / Express when the team's in JS; Spring where Java's mandated. Comfortable writing the boring plumbing other people won't.
AWS, GCP, Azure — worked in all three, favor GCP for ML tooling. Docker and CI/CD as table stakes. HPC clusters for training the genuinely large things; Linux everywhere else.
PostgreSQL, MongoDB, MySQL. Pandas and NumPy in my sleep. ETL design experience spanning insurance claims to plant-floor telemetry — which is to say, both the clean kind and the kind that comes in at 3am.
PLC and SCADA systems, IoT sensor networks, edge compute, LiDAR, predictive maintenance. The career before this career — and the reason I care so much about the part of AI that has to actually run.
Where I trained.
Certifications
- AWS Solutions Architect — Associate
- MSFT Azure Fundamentals
- IBM Data Science Professional
- GOOGLE IT Support Specialization
- UDA Data Analyst Nanodegree
- LPI Ubuntu Linux Professional
- 6σ Six Sigma Green Belt
- AI AI for Product Management
Let's build
something that runs.
Open to AI / ML engineering roles in France and remote-anywhere. Talent Passport visa in hand, immediately available. If you need computer vision, LLM systems, or the industrial-grade kind of AI that actually has to work — reach out.
jek283@pitt.edu↗