👋 About me

I'm a backend developer who builds the APIs and services behind products — and lately, the AI that runs inside them: RAG pipelines, LLM agents and computer vision models.

• 🔭 Currently building an ASP.NET Core Web API for a finance & invoicing app (internship)
• 🧠 I ship RAG systems and MCP servers, and fine-tune my own models when an external API isn't the right fit
• 🏆 TechXtile Challenge Level-Up finalist · TEKNOFEST competitor (UGV control station, AI in Aviation) · MTA AI Specialization capstone: multi-sensor anomaly detection on UAV telemetry
• 💬 Don't feel like reading? Chat with my AI clone instead

⚙️ Backend

• Finance & invoicing API (internship · private) — ASP.NET Core Web API serving a Flutter app: database-first EF Core on SQL Server, JWT authentication on top of ASP.NET Identity.
• The APIs behind my AI projects — an ASP.NET Core Minimal API with SSE streaming (foundry-local-rag-agent), a FastAPI service on Hugging Face Spaces (Portfolio) and a Dockerized MCP server (story-to-music).

🧠 LLM & RAG

BUĞRA.AI — Portfolio · live ↗
RAG-powered AI avatar: visitors chat with my digital clone instead of reading a CV. Hybrid retrieval (multilingual embeddings + BM25, fused with RRF) checked by an offline eval suite.
FastAPI LangChain ChromaDB Groq React Three.js

foundry-local-rag-agent
100% on-device agentic RAG: chat with Word/PDF/Excel files with source citations and generate Word/Excel/PDF reports. Built during the Microsoft Foundry Local Summer School.
.NET 10 ASP.NET Core Semantic Kernel Foundry Local SQLite

story-to-music · demo ↗
MCP server that turns Turkish stories into Suno/Udio prompts and structured lyrics with two fine-tuned mT5 models — no external API.
Python MCP mT5 Hugging Face Docker

👁️ Computer Vision

SFDDS · 🏆 TechXtile Challenge Level-Up finalist
Textile defect detection on the ZJU-Leaper dataset with YOLOv7, YOLO11 and YOLO26, exported to ONNX for real-time use.
PyTorch Ultralytics OpenCV ONNX

TURANOVA · TEKNOFEST 2026 AI in Aviation (private)
Autonomous image processing system: YOLO11 object detection, visual odometry and DINOv2-assisted cross-modal reference matching.
YOLO11 DINOv2 Visual Odometry

Brain Tumor MRI Classification · Akbank Deep Learning Bootcamp
Four-class brain tumor classification on ~7,000 MRI scans with a CNN baseline and EfficientNetB0 fine-tuning, explained with Grad-CAM.
TensorFlow Keras Grad-CAM

📱 Mobile

Sağlıkla
AI health companion for diabetes and celiac patients: barcode scanning via OpenFoodFacts, Gemini-powered food analysis from photos, meal and hydration tracking.
Flutter Firebase Gemini

Vakit
Prayer times and Qibla compass app for Android.
Kotlin Jetpack Compose Room Retrofit

🛠️ Tech stack

Backend	ASP.NET Core · EF Core · SQL Server · REST · SSE · JWT
AI / ML	LangChain · Semantic Kernel · Hugging Face · Ultralytics YOLO · ChromaDB
Mobile	Bloc · Jetpack Compose · Room · Retrofit

📈 GitHub activity