AI & Computer Vision Development
I build AI software that understands camera input in real time and turns it into action: target-UAV detection with Faster R-CNN on KAPGAN, helipad detection with YOLOv3-tiny in a precision-landing system, and natural-language commands turned into autonomous flight with a language model (Gemma 3) on Atmaca. The focus is not a demo but a system that runs reliably in the field.
First step: Discovery & Feasibility — Free
Problems it solves
- You need to detect and track an object in a camera feed in real time.
- Off-the-shelf models do not recognise your object, so a model has to be trained on your data.
- Detections must reliably drive an action — a control command, an alert or a report.
- You want to connect a language model (LLM) to your system without leaving decisions to guesswork.
- The model has to run locally (on a device or your own server) so data never leaves.
What you get
- A custom object-detection model: data preparation, training and evaluation
- A real-time image-processing pipeline with OpenCV
- Integration into your system (Python, C# / .NET, MAVLink or an API)
- LLM and MCP (Model Context Protocol) integration: a deterministic core plus an explaining AI
- Source code, model files and documentation
How we work
- 1
Free discovery call
I listen to your need, your current system and your goals, and we talk openly about what is technically feasible.
- 2
Scope and quote
A written scope, milestones, an exact price and a timeline. No surprise costs — what is included is clear from the start.
- 3
Build and regular demos
We first validate the approach on your data with a small proof of concept, then grow the system step by step.
- 4
Test and handover
Testing with real and hard-case samples; we review the results together and I hand over the source code and model.
Tech stack
Related projects
Real projects I have built in this area.
Teknofest 2021 · Turkey Champion
KAPGAN Combat UAV
National champion and winner of the "Most Original Software" award at Teknofest 2021 Combat UAV Competition — I was the sole developer of all three software units.
My first international freelance job · Upwork/Fiverr
Vision-Based Autonomous Precision Landing
A precision-landing system that uses AI to detect a helipad "H" marking and lands the drone exactly on its centre — continuously re-centring during descent so wind/drift cannot throw it off. My first serious international freelance job.
Open source · R&D
Atmaca — LLM-Powered Autonomous UAV
An experimental UAV you command in plain (natural) language; a language model (Gemma 3) interprets the commands and turns them into autonomous flight.
Open source · MCP server
ardupilot-mcp — Flight Log Diagnosis Tool
An open-source MCP server that reads an ArduPilot flight controller's ".bin" black-box log and explains in plain language why a drone crashed or flew badly — grounded in a deterministic check engine, not AI guesswork.
Teknofest 2022 · Finalist (9th)
Armed Autonomous Strike UAV
Teknofest 2022 TÜBİTAK Open-Mission UAV finalist: a hexacopter that runs the entire mission — from takeoff to strike — autonomously. I coded its "brain", the main management unit.
Frequently asked questions
How much does an AI / computer vision project cost?
The price depends on the scope; there is no fixed price list. The easiest way to start is Discovery & Feasibility: requirements analysis, technical roadmap, rough plan and proposal document. Discovery and feasibility are free. The price and timeline of the project itself depend on its scope and come as a firm quote at the end of that work.
How much data is needed to train the model?
It depends on how varied the object is and on the environment. In the discovery call we review the data you have and, if needed, plan data collection and labelling.
Does the model run in the cloud or on the device?
Whichever fits. For example, the precision-landing system runs its YOLOv3-tiny model locally through the OpenCV DNN module, and Atmaca runs its language model locally with Ollama. Local inference is preferred when data must not leave your premises.
Can you add AI to my existing software?
Yes. I connect the model to Python or C# / .NET systems as a library or a service, and turn its output into a control command, an alert or a report.
Do you integrate LLMs?
Yes. On Atmaca I turned natural-language commands into autonomous flight with Gemma 3; in ardupilot-mcp a language model explains the findings of a deterministic check engine in plain language. I prefer leaving decisions to a verifiable core, not to the model.
Do you work remotely, and in which languages?
Yes. I am based in Türkiye (GMT+3) and work remotely with clients worldwide; meetings are held in Turkish or English and scheduled around your time zone.
Let's talk about your project
We clarify your need together in a short call; then I send the scope, exact price and timeline in writing.