I build projects end to end and ship them with tests, CI and documentation. Recent work covers a retrieval-augmented Q&A tool for codebases, a reinforcement learning agent trained on a live Minecraft server, a network intrusion detection system, an AI study platform in production and a log-analysis tool published on PyPI.

• Third-year Computer Engineering student (2024 – 2028)
• Interested in AI/ML, cybersecurity, networking and developer tools
• Most recent project: RepoLens AI, which answers questions about a GitHub repo and cites the exact files and lines

Paste a GitHub URL and ask questions about the code. Answers cite file paths and line ranges, every citation is verified against the code the model was actually shown, and questions the repo can't answer get refused instead of invented. Retrieval combines embeddings and BM25 (merged with reciprocal rank fusion), followed by a cross-encoder reranker. On 71 labelled questions over pallets/flask, hybrid search reaches 98.6% Recall@3 and 100% Recall@5, and all 18 of 18 questions about things Flask doesn't have were refused. Code is chunked along function and class boundaries with Tree-sitter, re-indexing only re-embeds changed files, and an agent mode follows names across files with read-only tools. Runs with one docker compose up, with CI on every push.
Python FastAPI React TypeScript Qdrant Tree-sitter sentence-transformers Gemini API Docker

Upload course material and get AI summaries, generated quizzes and an adaptive study plan that gives more time to the topics you score poorly on. Includes spaced-repetition reminders, JWT auth, and a Jest/Vitest test suite running in GitHub Actions.
React Node.js Express MongoDB Gemini API GitHub Actions

A Minecraft bot that joins a Java server as a player and takes commands (chop trees, craft tools, mine ore), plus a Gymnasium environment for training it with behaviour cloning and PPO over a WebSocket bridge. In evaluation, the shared mining policy collected 2.75 more ore per episode than a random baseline (95% CI [0.53, 4.97]). Supports optional natural-language commands through an LLM.
Python Gymnasium PPO Node.js Mineflayer WebSocket

Real-time intrusion detection system with four sliding-window detectors (port scan, brute force, traffic spike, SYN flood). Alerts are stored in SQLite and streamed to a live Flask dashboard. Runs on live traffic via Scapy, or in a simulation mode that works without root.
Python Scapy Flask SQLite pytest

Zero-dependency CLI for log analysis, installable with pip install loglynx. Scores log health, clusters recurring errors, flags time windows that break from the file's own baseline and compares runs to catch regressions after a deploy. Also offers live watch, HTML/JSON reports and a dashboard. Released to PyPI through GitHub Actions.
Python CLI Anomaly Detection GitHub Actions PyPI

Languages: Python · JavaScript · TypeScript · C# · SQL
Backend: Node.js · Express · FastAPI · Flask
Frontend & mobile: React · Vite · Capacitor
Data & AI: RAG · hybrid search (BM25 + embeddings) · Qdrant · sentence-transformers · Gemini API · Gymnasium · PPO · PostgreSQL · MongoDB · SQLite
Security & networking: Scapy · intrusion detection · TCP/IP fundamentals
Tooling: Git · Docker · GitHub Actions · PyPI publishing · Jest · Vitest · pytest · Linux

• B.Sc. Computer Engineering, Mersin University (2024 – 2028)
• Huawei ICT Academy: Computer Networks V1.0 and HCIA-Datacom V1.0 (course completion, 2026)