Computer Engineering student at Gazi University, building production-minded software at the intersection of AI systems and full stack product engineering.
- Software Engineering — Clean architecture, OOP and design patterns, data structures & algorithms, and maintainable codebases across Python, TypeScript, Java, C#, C and Rust.
- AI / ML — LLM-driven applications, speech pipelines (STT → LLM → TTS), prompt and context engineering, and applied ML with the Python data stack.
- Full Stack Development — React/TypeScript frontends, Flask/Node backends, REST APIs, and browser extensions shipped end to end.
- Product Engineering Mindset — I ship tools people actually use: open-source, documented, licensed, and iterated in public.
Open To
Software Engineering Internships · AI / ML Engineering Roles · Open Source Collaboration · Research Projects
Languages
Frontend
Backend & Databases
Cloud, DevOps & Tooling
CADLift — Prompt & Drawing to Production-Ready 3D Geometry
Open-source engine that converts natural language prompts, images and 2D DWG/DXF drawings into production-ready 3D geometry, compatible with AutoCAD, FreeCAD, Blender and Unity.
| Stack | |
| Scale | Multi-format CAD interoperability across four major 3D toolchains |
| Performance | Deterministic geometry pipeline built on reusable parametric primitives |
| Security | MIT licensed, local-first processing — no proprietary drawing upload required |
| Impact | Most-starred repository in my portfolio; a bridge between AI and engineering CAD workflows |
| Repository | github.com/Vartmor/CADLift |
Engineering CAD work is repetitive at the modeling stage. CADLift attacks that bottleneck by treating geometry as a compilation target: the input can be a prompt, a raster image or a legacy 2D drawing, and the output is clean, editable 3D geometry that drops straight into an existing engineering pipeline.
GaziAI Avatar — Real-Time Conversational 3D Avatar
Interactive AI-driven 3D avatar with a Flask backend and a React / React Three Fiber frontend, wired to a real-time speech-to-text → LLM → text-to-speech loop.
| Stack | |
| Scale | Full voice conversation loop rendered against an animated 3D character |
| Performance | Streaming STT/TTS stages to keep perceived latency conversational |
| Security | MIT licensed; API keys stay server-side and are never exposed to the browser |
| Impact | End-to-end multimodal AI system: audio in, reasoning, audio and animation out |
| Repository | github.com/Vartmor/GaziAI-Avatar |
The hard part of a talking avatar is not the model, it is the orchestration: capturing audio, transcribing it, reasoning over it, synthesizing speech and driving the mesh — all without the user feeling the seams. This project is my study of that pipeline.
BringYourSub — AI Subtitles for YouTube, Bring Your Own Key
Free and open-source browser extension that generates AI subtitles for YouTube videos using the user's own OpenAI API key, with context-aware translation and a Whisper fallback when no native transcript exists.
| Stack | |
| Scale | Cross-browser support for Chromium and Firefox |
| Performance | Full-transcript context passing for coherent translation instead of line-by-line drift |
| Security | Bring-your-own-key model — credentials stay on the user's machine; MIT licensed |
| Impact | Makes any YouTube video watchable in any language without a paid subtitle service |
| Repository | github.com/Vartmor/BringYourSub |
Line-by-line machine translation loses the thread of a video. BringYourSub sends the full transcription as context so the translation stays coherent across the whole runtime, and falls back to Whisper when YouTube has nothing to offer.
MindBridge — AI-Assisted Mental Health Companion
Modern, AI-assisted mental health companion app for therapy patients: emotion tracking, journaling, task management, relaxation exercises and a safety-conscious AI chat.
| Stack | |
| Scale | Five integrated modules under one patient-facing product surface |
| Performance | Responsive, low-friction daily check-in flows backed by local state |
| Security | Safety-guarded AI chat boundaries; sensitive personal data treated as a first-class concern |
| Impact | Built as a companion between therapy sessions — never a replacement for clinical care |
| Repository | github.com/Vartmor/MindBridge-Psychology-App |
Building software adjacent to a clinical domain forces a different engineering posture: guardrails before features, and a product that always points back to a real therapist.
RataPulse — Terminal Process Monitor in Rust
Your system's pulse, right in the terminal — an htop-style Linux process monitor written in Rust with Ratatui.
| Stack | |
| Scale | Live process table sourced directly from the Linux /proc filesystem |
| Performance | Zero-GC Rust core with a redraw loop designed for a constant refresh budget |
| Security | Memory-safe by construction; read-only system introspection; MIT licensed |
| Impact | Systems-programming deep dive — phase 1 of a 5-phase roadmap, in active development |
| Repository | github.com/Vartmor/RataPulse |
A deliberate exercise in low-level engineering: ownership, lifetimes, /proc parsing and terminal rendering, with no runtime and no garbage collector to hide behind.
UniversiteDersSecim — Course Schedule Combination Engine
Schedule builder that generates and filters university timetable combinations, so course registration day stops being a brute-force search.
| Stack | |
| Scale | Combinatorial generation across the course catalog under user-defined constraints |
| Performance | Conflict-aware pruning keeps the candidate space tractable |
| Security | GPL-3.0 licensed and fully client-side — no student data leaves the browser |
| Impact | Turns an annual manual scheduling headache into a filter-and-pick workflow |
| Repository | github.com/Vartmor/UniversiteDersSecim |
Constraint satisfaction with a friendly face: the user states preferences, the engine enumerates only the schedules that survive them.
Smart Delivery System — OOP & Design Patterns in Java
Task-processing system for smart delivery, built to exercise object-oriented design end to end: Strategy pattern, generics, polymorphism and custom exceptions.
| Stack | |
| Scale | Pluggable delivery strategies, swappable at runtime without touching call sites |
| Performance | Type-safe generic task pipeline — no reflection, no casting |
| Security | Custom exception hierarchy for explicit, non-swallowed failure handling |
| Impact | Reference implementation of clean OOP architecture in my coursework portfolio |
| Repository | github.com/Vartmor/AkilliTeslimatSistemi |
Written as a proof that design patterns are not academic decoration: swapping a delivery strategy should be a one-line change, and this codebase makes it one.
| Contribution | Details |
|---|---|
| Maintainer — CADLift | MIT-licensed AI-to-CAD engine, the most-starred project in my portfolio |
| Maintainer — BringYourSub | MIT-licensed cross-browser extension for AI-generated YouTube subtitles |
| Maintainer — GaziAI Avatar | MIT-licensed real-time STT → LLM → TTS conversational avatar |
| Maintainer — RataPulse | MIT-licensed Rust terminal process monitor, in active development |
| Community | Stack Overflow contributor |
Language Distribution
| Language | Share | |
|---|---|---|
58.1% |
███████████████████████ | |
33.5% |
█████████████ | |
3.2% |
█ | |
2.0% |
█ | |
1.4% |
█ | |
1.4% |
█ |
learning:
- Rust systems programming and terminal UI engineering
- LLM orchestration, evaluation and context engineering
- Distributed systems and cloud architecture fundamentals
building:
- CADLift — AI-driven prompt/image/DWG to 3D geometry
- RataPulse — htop-style process monitor in Rust (phase 1 of 5)
- MindBridge — AI-assisted mental health companion
exploring:
- Multimodal AI pipelines (speech, vision, geometry)
- MLOps, model serving and inference cost optimization
- Open-source developer tooling
open_to:
- Software Engineering Internships
- AI / ML Engineering Roles
- Open Source Collaboration
- Research Projects

