Bridging applied research with field-ready deployment, low-latency systems, and AI security.
I work at the intersection of applied AI research and forward-deployed engineering. I combine systematic experimentation and competitive modeling with the discipline needed to deploy, optimize, and scale ML systems in production and on edge devices.
Currently an independent AI Solutions Consultant (since 2024). Previously ML R&D Engineer at Datamarkin, where we built the ML infrastructure of a no-code computer vision platform from zero.
- Computer Vision & Applied Research: Deep learning pipelines for visual inspection, detection, segmentation, pose analysis, and physiological signal extraction (rPPG).
- Forward Deployment & Systems: Turning problem statements into production-ready AI services, rapid PoCs, and scalable API-driven architectures.
- Model Optimization & Edge AI: Quantization, ONNX, and TFLite for lower latency and memory use, including deployment on Raspberry Pi / IoT devices.
- Applied GenAI: LLM inference with vLLM (latency–throughput trade-offs), RAG prototypes with vector retrieval and prompt routing, and RLHF preference/safety data work.
- AI & Cybersecurity: Deep learning for intrusion detection on 5G traffic, LLM red-teaming, and security-minded design for IoT and multi-tenant ML platforms.
| Writeup | Competition | Summary |
|---|---|---|
| GemmAid: Multilingual Emergency Mobile Triage with Gemma 4 | Gemma 4 Good Hackathon | Rapid-deployable, locally running crisis triage: no cloud and no setup, just Gemma 4 and a laptop. |
| Health Companion | MedGemma Impact Challenge | Multilingual AI triage with explainable clinical intelligence, built as a bridge between patient and physician. |
| Would Your AI Save You? Loyalty and Sacrifice in LLM Moral Choices | OpenAI gpt-oss-20b Red-Teaming | Red-teaming study of how an LLM handles moral choices involving loyalty and sacrifice. |
| Project | What I built | Stack |
|---|---|---|
| Datamarkin (No-code / End to end Computer Vision platform) | Modular microservices (Trainer, Inference, SQL Wrapper) behind a REST gateway; hybrid on-prem/cloud infra with multi-tenant isolation. | PyTorch, TensorFlow, ONNX, OpenCV, Flask, Detectron, SAM |
| Obee (Smart agriculture IoT PoC) | Field-sensor telemetry over MQTT/TCP to a central unit, plus edge-optimized TFLite models on Raspberry Pi for real-time, offline crop disease detection. Eclipse Open IoT Challenge Semi-Finalist. | Python, TensorFlow Lite, MQTT, SQLite, PyQt |
| 5G Intrusion Detection (MSc research) | Real-time detection on the 5G-NIDD dataset: a Random Forest selects features for a Deep Neural Network (fDNN), reaching high accuracy with minimal false positives. | Python, Scikit-learn, Pandas |
| Area | Tools |
|---|---|
| Modeling & Research | PyTorch, TensorFlow, OpenCV, Scikit-learn, Hugging Face Transformers |
| Serving & MLOps | Docker, FastAPI, Flask, ONNX Runtime, Weights & Biases, MLflow, Triton |
| Cloud & Systems | AWS, GCP, Azure, Linux administration, TCP/IP |
| Data & Search | Pandas, NumPy, Matplotlib, SQLite, Pinecone |
| Edge & IoT | Raspberry Pi, MQTT, RTSP |
| Languages | Python, Bash |
- 🏅 Kaggle Competitions Expert: Silver Medal (LLM - Detect AI Generated Text), Bronze Medal (LANL Earthquake Prediction)
- 🏆 Eclipse Open IoT Challenge: Semi-Finalist
- 🎓 MSc, Cyber Security, BS, Electronic & Telecommunication Engineering
- 📜 NVIDIA AI Expert · DeepLearning.AI MLOps · IBM Data Science & AI Engineering · Google Cloud Data Engineering & ML
I'm open to conversations about applied computer vision, edge deployment, LLM systems, and AI consulting. Reach me on LinkedIn.