Skip to content
View absoyak's full-sized avatar
🎯
Focusing
🎯
Focusing
3 followers · 2 following

Block or report absoyak

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
absoyak/README.md

Burak Soyak

Senior Computer Vision & Applied AI Engineer working on real-time sports and broadcast systems. I build GPU-accelerated vision pipelines that run in live environments, not just on clean datasets.

  • Chief Timing Officer, F1H2O & Aquabike World Championships: 50+ international events operated on-site
  • Imperial College London, Professional Certificate in ML & AI (2026): 100 % on all assessed work

AdSwapAI: original broadcast frame vs AI-replaced boards
AdSwapAI: pitch-side boards replaced by AI, players untouched

What I build

  • Detection, segmentation and tracking pipelines: Detectron2, YOLO, Faster/Mask R-CNN, Meta SAM 2/3
  • GPU inference: PyTorch CUDA, TensorRT, batched and asynchronous processing
  • Live video pipelines: NDI, SRT, FFmpeg, broadcast-grade latency
  • Bayesian optimisation with Gaussian-process surrogate models
  • Race timing and sports data systems for international live events

Projects

Project What it is
AdSwapAI · adswap.ai AI replacement of pitch-side advertising boards in football footage. Custom Detectron2 Mask R-CNN board segmentation with player/ball protection, Flask + nginx + Docker demo; now moving to SAM3 prompt-based detection. The repository carries the whole R&D history since Jan 2025, step by step.
F1H2O AI Timing Faster R-CNN boat detection integrated into official race timing; outperformed manual timing in field testing. Production system, private.
LapData F1H2O race-intelligence stack on real championship data: Leaflet circuit visualisation, SQL Server telemetry schema, FastAPI backend, LangChain agent with SQL + RAG + prediction tools, local LLM via Ollama.
GP Optimisation Capstone Imperial ML/AI capstone: black-box optimisation (F1–F8, 2D–8D) with Gaussian processes and custom acquisition strategies.
Computer Vision Playground Production-minded experiments: real-time video pipelines, pose, tracking, GPU benchmarks.
ndi-jpeg-capture NDI stream capture at a fixed FPS with tick-precise timestamps (.NET).

Stack

Python · C# / .NET · PyTorch · OpenCV · Detectron2 · SAM3 · CUDA · TensorRT · FFmpeg · NDI · Docker · AWS · SQL Server · FastAPI · LangChain · Ollama · SQLAlchemy

Links

LinkedIn in/absoyak · iocode.co.uk · adswap.ai

Popular repositories Loading

  1. Computer-Vision-Playground Computer-Vision-Playground Public

    Production-minded computer vision experiments: real-time video pipelines, pose, tracking and GPU benchmarking.

    Python 3

  2. imperial-ml-ai-capstone imperial-ml-ai-capstone Public

    Imperial ML/AI Capstone

    Python 1

  3. adswapai adswapai Public

    AI-based replacement of pitch-side advertising boards in sports video: Detectron2 Mask R-CNN, Flask, Docker, plus the full R&D history

    Python

  4. absoyak absoyak Public

    Software developer building an ML/AI portfolio (computer vision, applied ML, and production systems).

  5. ndi-jpeg-capture ndi-jpeg-capture Public

    Captures an NDI video stream at a fixed FPS, saving frames as JPEG files named with .NET DateTime.Ticks for precise timestamp-based retrieval.

    Python

  6. lapdata lapdata Public

    AI agent for motorsport timing data natural language queries, race prediction, and performance analytics. Built on real F1H2O Championship infrastructure.

    JavaScript