M.Sc. Information Engineering @ Technical University of Munich (TUM), Heilbronn
I build AI systems that run where the data lives. Most of my recent work sits at the intersection of retrieval and low-level performance: indexing and searching large vector collections, compressing models so they fit on constrained hardware, and writing CUDA kernels to make them fast.
Before TUM I completed a B.Sc. in Computer Engineering at Cairo University (1.2 German scale, Distinction with Honors).
Open to Werkstudent roles in AI/ML or software engineering — Heilbronn, Stuttgart, or remote. Available immediately, 20 h/week during the semester.
AI / ML — PyTorch · Hugging Face Transformers · RAG · ChromaDB · vector databases · IVF & product quantization · model compression · OpenCV · BERT · YOLOv5
Systems — C · C++ · CUDA · GPU computing · parallel programming · Embedded C · AVR · ARM Cortex-M
Web & Backend — Python · TypeScript · React · Next.js · Node.js · Flask · REST APIs · PostgreSQL · MySQL · MongoDB
Tools — Git · Linux · Docker
Inferex — B.Sc. graduation project. A privacy-preserving desktop assistant that builds a searchable memory of your computer activity: it captures what you read and watch, indexes it in a vector database, and answers natural-language questions about it — entirely offline. Includes a custom CUDA kernel for lossless LLM compression that cut Qwen3-4B's VRAM from 8.4 GB to 5.8 GB with no accuracy loss, enough to run on a 6 GB consumer GPU.
DBMS-CUDA — A database engine that offloads SQL execution to the GPU, parsing DuckDB execution plans and streaming batches beyond GPU memory. 3.8x faster joins than the CPU baseline.
Almost Google — A search engine built from scratch: crawler indexing 6,009 pages, inverted index, ranking, and phrase search, with a full frontend and backend.
On The Go — Full-stack restaurant management system. React frontend, REST API, PostgreSQL schema, with authentication and role-based permissions across customer, employee, and admin users.
EntranceSystem — License plate detection and recognition. OpenCV pipeline with edge detection, morphological operations, and contour analysis, then OCR via HOG features and logistic regression, served through Flask.
ATmega328P-Drivers — Peripheral drivers written from the register level up: GPIO, timers, I2C, SPI, UART, ADC, keypad, and seven-segment display.
- LinkedIn: mostafaelsayed2002
- Email: mostafa.elsayed.2002@gmail.com