Applied AI systems for computer vision, edge inference, audio understanding, LLM applications, and measurable deployment workflows.
AI Engineering is the central track in my career. I focus on turning models into usable systems: collecting or preparing data, selecting baselines, evaluating failure modes, integrating inference into applications, and documenting deployment constraints. My current work is strongest in computer vision, audio AI, and edge inference.
- Computer vision and real-time video analytics
- Edge AI on Raspberry Pi, Hailo, mobile devices, and embedded Linux
- Audio classification, speech-to-text, and text-to-speech pipelines
- LLM applications, retrieval-augmented generation, and safety-aware product design
- Azure Machine Learning workflows and model registration
- YOLO, ONNX, TensorFlow Lite, OpenVINO, and hardware-aware benchmarking
- RTSP ingestion, zone analysis, fall detection, fire/smoke detection, and event generation
- Reproducibility, test coverage, observability, and privacy-aware demos
| Project | Scope | Evidence to publish | Status |
|---|---|---|---|
| zone-analysis | C++ RTSP person detection, polygon zones, dwell-time alerts, snapshots, and CSV events | Demo video, reconnect test, FPS/latency table | In Progress |
| hailoai | Raspberry Pi 5 CPU vs Hailo NPU RTSP benchmark | Reproducible benchmark table, thermal chart, model matrix | In Progress |
| baby-cry-ai-detector | Two-stage cry detection and reason classification with honest class-level evaluation | Public-safe sample, model card, inference demo | In Progress · Private |
| azureml-baby-cry-classification | Azure ML audio preprocessing, CNN training, evaluation, and model registry workflow | Existing metrics plus reproducibility check | Completed |
| Flutter YOLO11n Fire & Smoke Detection | Mobile on-device YOLO/TFLite inference | Device FPS, model size, demo capture | In Progress |
| GM_Voiceai | Local-first Turkish speech assistant pipeline | Supported-command matrix, latency, audio demo | In Progress |
| emotional-support-ai | Safety-aware LLM chat MVP with FastAPI and PostgreSQL | Evaluation set, refusal behavior, privacy notes | In Progress · Private |
| safeguard-edge | Edge safety monitoring across inference, backend, dashboard, and mobile layers | Sanitized architecture and end-to-end demo | In Progress · Private |
openvino-rtsp-benchmark |
Planned CPU/iGPU benchmark with reproducible latency and throughput measurements | Benchmark harness, environment lockfile, results table | Planned |
multi-camera-fall-detection |
Planned temporal fall-detection pipeline with privacy-safe test video | Baseline comparison, false-positive review, event API | Planned |
rag-evaluation-workbench |
Planned document RAG service with retrieval and answer-quality evaluation | Dataset, metrics, FastAPI service, test report | Planned |
- Publish measured results and privacy-safe demos for
zone-analysisandhailoai. - Add model cards, dataset provenance, limitations, and reproducible setup instructions to audio projects.
- Create this landing repository and link every featured project back to it.
- Start the OpenVINO RTSP benchmark with a CPU baseline.
- Complete one multi-camera or temporal vision project with an event API and automated tests.
- Publish a small RAG evaluation workbench with retrieval metrics and traceable sources.
- Standardize CI, linting, release notes, Docker support, and benchmark templates across featured projects.
- Maintain three production-style AI reference systems: real-time vision, audio intelligence, and LLM/RAG.
- Compare at least three edge inference backends using the same documented evaluation protocol.
- Publish technical articles that connect architecture decisions to measured system behavior.
- Languages: Python, C++, SQL, Dart, TypeScript
- ML: PyTorch, TensorFlow, scikit-learn, YOLO, ONNX, TensorFlow Lite
- Vision & Audio: OpenCV, RTSP, Librosa, mel spectrograms
- Cloud & MLOps: Azure Machine Learning, Docker, GitHub Actions
- Edge: Raspberry Pi 5, Hailo AI HAT+, OpenVINO
- Applications: FastAPI, PostgreSQL, Redis, React, Flutter
| Working title | Project connection | Status |
|---|---|---|
| Benchmarking RTSP Inference on Raspberry Pi 5: CPU vs Hailo NPU | hailoai |
Planned |
| From Detection to Events: Building Dwell-Time Zone Analytics in C++ | zone-analysis |
Planned |
| Baby-Cry Classification: Why Class-Level Data Quality Matters | Audio AI projects | Planned |
| Evaluating RAG Beyond a Demo | rag-evaluation-workbench |
Planned |
Medium links will be added after publication.
This portfolio supports applications for:
- AI Systems Engineer
- Machine Learning Engineer
- Computer Vision Engineer
- Edge AI Engineer
- Applied AI Engineer
- MLOps Engineer with model-deployment responsibilities
The intended evidence is end-to-end engineering: model evaluation, real-time integration, deployment constraints, testing, and clear technical communication.
- Completed: reproducible scope delivered and documented
- In Progress: working scope exists, but evidence or engineering hardening remains
- Planned: proposal only; not presented as completed work
Portfolio status: In Progress