aelbikha.me / loadingread
I build research assistants, browser agents, and educational AI: the retrieval pipelines, tool execution, evaluation harnesses, and interfaces that make them useful.
B.S. in Computer Science from CUNY John Jay; starting an M.S. in CS · Machine Learning at Georgia Tech.
the seam between model and interface
An ML-and-systems engineer is most useful where the model meets the interface. I work across that seam: research, engineering, and design, to ship things people actually use.
Production retrieval-augmented generation. Hybrid vector / graph / keyword search, re-ranking, eval harnesses, and prompt-injection defence at the agent layer.
Type-safe, performance-budgeted apps with Next.js and FastAPI. WebSocket streaming, Postgres + Redis, JWT/RBAC, Dockerised deploys, 500+ concurrent users.
Applied machine-learning research: contrastive MPNet fine-tuning, XGBoost detection, and disjoint evaluation splits for prompt-injection intent.
Readable, responsive surfaces for complex systems: replayable browser workflows, approval states, tool calls, and the small decisions that make technical products feel clear.
a working log
Six years of work, school, and side projects, indexed by date. A working log of the systems, interfaces, and research that shaped the practice.
Graduate study specialising in machine learning: deep learning, NLP, and large-scale systems.
Built a tutoring agent that retrieves course content and routes student questions to relevant tools. Tuned retrieval and re-ranking for students with learning differences; added fine-tuned sentence-transformers, Redis caching, batching, and response streaming.
Built a quiz-generation pipeline with ChromaDB retrieval and Cohere re-ranking; integrated GPT-4o chat with WebSocket streaming in student-facing applications. Developed Dockerised FastAPI, PostgreSQL, Redis, React, and Flutter systems with RTL localisation across four languages.
Undergraduate degree covering computer science fundamentals, information security, and applied machine learning.
15 minutes · direct
Pick a time that works. The calendar updates with current availability.
Open booking page ↗