An AI engineer building LLM-powered products — RAG, agents and assistants — on top of four years of shipping computer vision to production.
Get in TouchI’ve spent nearly 4 years building computer vision systems that run in production — real-time object detection, tracking, OCR, and anomaly detection — taking projects from dataset design all the way through edge deployment on Jetson with TensorRT and ONNX, and integrating with industrial cameras and PLCs. I write clean, maintainable Python (OOP, SOLID) and care about getting the architecture and data pipeline right before touching the model. I’ve delivered a commercially licensed product and led a project that’s been running in production for over a year without retraining. I also mentor junior engineers and interns. I’ve now moved to the LLM team, where I bring the same production mindset to large language models, RAG pipelines and agentic systems.
Python
PyTorch
TensorFlow
OpenCV
YOLO
RT-DETR PaddleOCR
TensorRT
ONNX CUDA NVIDIA Jetson
PyQt6
FastAPI
Docker
MongoDB
PostgreSQL
Azure DevOps
Linux
LangChain2025 – Present
From a TÜBİTAK R&D prototype to a production fleet system deployed to Ankara Metropolitan Municipality. Accuracy raised to 87%; 10-vehicle rollout in progress.
Project details
Jan 2026 – Mar 2026
An end-to-end cobot application that turns photos or line art into physical pen drawings with a Lexium cobot, with a safety-interlocked desktop HMI for industrial demos.
Project details
Jan 2026 – Mar 2026
A standalone desktop platform covering the full object detection lifecycle — camera capture, annotation, training and real-time GPU-accelerated inference — in a single application.
Project details
Oct 2024 – Dec 2024
Eliminated missed-cell errors in cell counting and raised small-object detection accuracy to 95%+ with SAHI. Delivered as a subscription-based, server-licensed commercial product.
Project details
Apr 2024 – Jun 2024
Owned dataset design, class taxonomy and model training end-to-end, reaching 95%+ accuracy. Running in production for over a year without retraining.
YOLOv8 · Watchdog · Queue system · REST API
Jun 2024 – Aug 2024
Replaced a previous vendor’s high false-positive solution with a retrained model on existing cameras and a more stable desktop UI; handled on-site installation and user training.
OpenCV · PyQt5 · USB camera · PLC
Oct 2023 – Dec 2023
Dataset preparation, class design and model training for classifying weld defects in industrial X-ray images, meeting the client’s acceptance criteria for production.
CNN · Image classification · X-ray imaging
Apr 2025 – Jun 2025
Added a Jetson–Arduino trigger for spray-marking control to a field-deployed system and reduced frame latency in the video pipeline.
OpenCV · PyQt5 · Arduino · Jetson
Apr 2025 – Jun 2025
Literature-based drowsiness and distraction detection (eye and facial landmarks, head pose) with a redesigned real-time visual feedback layer.
OpenCV · MediaPipe · REST API
Jan 2024 – May 2024
Backend modules that evaluate factory-deployed models, filter mislabeled samples, generate ready-to-train model cards and queue training jobs automatically.
Python · AutoML · Training pipelines · API
Apr 2023 – May 2023
Improved presence/absence accuracy from ~80% to 95%+ by rebuilding the dataset and diagnosing a faulty camera setup that degraded image quality.
YOLO · Albumentations · Model benchmarking
May 2023 – Aug 2023
Corrected the label taxonomy and rebalanced the dataset, improving intact / defective / missing detection at the robotic pick-and-place point.
OpenCV · CNN/YOLO · Dataset creation
Practice projects built independently to develop hands-on LLM skills — not client-delivered or production systems.
2025 – Present
A RAG pipeline over sample documents using a FAISS vector database and LangChain / LlamaIndex to answer queries grounded in the source.
LangChain · LlamaIndex · FAISS · OpenAI API
2025 – Present
A multi-modal support assistant routing text, voice and image inputs through tools, and a multi-agent workflow for retrieval, analysis and reporting.
LangChain · Autogen · Function calling · Gradio
2025 – Present
An audio-to-summary tool comparing frontier and open-source models, and a price-prediction model fine-tuned with LoRA/QLoRA.
Transformers · PEFT · LoRA/QLoRA
Ithinka Information Technologies
Jul 2022 – Present · 3 roles
Oct 2023 – Present
Own end-to-end delivery of computer vision projects: dataset design, model training and optimization, edge deployment and production monitoring. Work directly with clients from requirements through go-live, including on-site installation and end-user training. Mentor junior engineers and interns, and produce technical documentation for every delivered project.
Oct 2022 – Sep 2023
Started on the data team, building an anomaly detection system for abnormal network traffic on monitored servers — including balancing a highly imbalanced dataset — and presenting weekly progress reports to stakeholders. Then moved into computer vision, delivering dataset and model-training work on industrial quality-control projects such as End-of-Line Inspection and Spring Sorting.
Jul 2022 – Sep 2022
Mezopotamya Living Lab · program delivered by Ithinka
Completed a 3-month data science training program covering data analysis and machine learning workflows, and was hired by Ithinka at the end of the program.
Erciyes University
2015 – 2018
LLM Engineering: Master AI, Large Language Models & Agents
Udemy · Aug 2025
Image Processing with Deep Learning: Python, OpenCV, Keras
Udemy
Python A-Z™: Data Science and Machine Learning
Udemy
Data Science Training
Mezopotamya Living Lab · delivered by Ithinka
Python and TensorFlow for Data Science
BTK Akademi
Have an LLM, RAG or computer vision project in mind — or something that needs to go from prototype to production? I’d be glad to hear about it. I work remotely.
Email: mucahit0yilmaz@gmail.com.
Or let’s connect on
Python · Ultralytics YOLO · OpenCV · ByteTrack · TensorRT · Jetson Orin Nano · MongoDB · GPS/GNSS · REST API · LTE · OTA
An end-to-end cobot application that converts photos or line-art images into physical pen drawings executed by a Lexium cobot.
Python · PyQt6 · OpenCV · NumPy · OpenAI API · Lexium Cobot · TCP/IP · Path planning · Desktop HMI
A standalone desktop platform covering the full object detection lifecycle in a single application, enabling local, project-based model development without separate labeling or training tools.
Python · PyQt6 · OpenCV · Ultralytics YOLO · PyTorch · CUDA · Albumentations · ONNX · QtCharts · Daheng GigE SDK
A quality control system that detects and evaluates each solar panel cell individually on the production line.
Python · OpenCV · YOLOv8 · SAHI · PyQt5 · MongoDB · SQLite · FastAPI