Software Engineer · Full-Stack · AI Systems
B.A.Sc. Artificial Intelligence @ Deggendorf Institute of Technology
I gravitate toward the role of technical anchor – the person who makes sure things actually ship, then keeps them running. Most of my professional experience is full-stack web (Next.js, Python, Docker); on my own time I build products people around me use and ML systems I can defend in a paper. I'm comfortable owning architecture decisions and writing the boring-but-critical infrastructure, though I've been actively practicing not over-engineering early – scoping to the simplest thing that proves the idea before the abstractions show up.
These are the ones I'd actually walk you through in an interview.
| Project | Stack | What it does |
|---|---|---|
| ☕ Kaffeelisten | React, TypeScript, Supabase, Vercel | Replaced the paper coffee sheet at ITC1 Deggendorf, a 28-company coworking campus – no login, under 15 seconds per entry, monthly PDF + Excel statements. Two-developer team; I own the data model, row-level security, billing and reporting, and keep it running in production. Started at the B4Y3RW4LD Hackathon. Pitch deck |
| 🩻 CheXVision | Python, PyTorch, DenseNet-121, Hugging Face, Kaggle | Chest X-ray pathology detection on 112,120 NIH images. A custom SE-ResNet from scratch (0.814 macro AUC-ROC) against DenseNet-121 transfer learning (0.846, vs the published CheXNet 0.841) under one fixed protocol. Architecture & cloud lead, team of four. Paper · live demo |
| 🏫 THD Room Finder | React, TypeScript, Vite, Capacitor | Which THD study rooms are free right now – and for how long. One codebase as an installable PWA and an Android app: duration-first ranking, 2.5D campus map, live EN/DE, framework-free unit-tested domain core. Used by students. |
| 🚢 Full-Stack ML Platform | React, FastAPI, PostgreSQL, Docker | Five-service product for auth, model training and inference, and per-user prediction history. Led the seven-person team: architecture, proposal, user stories, and a review on every PR. |
| 🗑️ Waste Sorting Assistant | Python, YOLOv8, PyTorch, Streamlit, Docker | Real-time bin detection for international students in Germany: YOLOv8 fine-tuned on 466 local images, 95.2% mAP@0.5. Three-person team; I owned DevOps and deployment. |
| 🌊 GKD4HYDRO (private client repo) | Python, pandas, SciPy, Parquet, pytest | Leading a four-person client project that turns Lower Bavarian river data into reproducible oscillation analyses – ramp rates, cycles, seasonal percentiles – from a single command. |
Other projects (earlier work, prototypes, learning projects)
- 🌿 Zelara – Edge-first app core: shared state, device linking, and conditional builds where inference stays on-device.
- 🚂 Deutsche Bahn Delay Predictor – 2.06 million DB records, 16 engineered features, three regression approaches compared on held-out data. Team of three.
- 🏥 Project Apero – Educational stroke-risk study app with a locally run Rasa assistant. Not a clinical product.
- 🤖 Devablos V2 – Real-time React/Node.js app with OpenAI over secure WebSockets; led the five-person first-year team.
- 🌍 iTravel – Android travel app with map discovery, GPS-validated reviews, and weather/photo feeds.
- 📄 Document Conversion Solutions – Format-conversion toolkit across Python APIs, a web frontend, and npm packages.
- 🦄 UNICorn – Pre-ChatGPT university assistant on GPT-3, Whisper, and semantic search.
- 🎮 WildForFood – Web-based Unity game built around rapid gameplay iteration.
- Voyager // 56K – this portfolio as a game: a rubber duck sailing the flooded internet of 1999, where every island is a project
- Leading GKD4HYDRO, a client project on river-level oscillations in Lower Bavaria
- Keeping Kaffeelisten running in production for the ITC1 campus
- Final stretch of my B.A.Sc. in AI at THD
- A twelve-month, build-led robotics programme: control, state estimation, electronics, CAD
Let's talk → alex.rudaev@outlook.com