ML Engineer who goes deep from training pipelines and model deployment to systems-level work in Rust and CUDA. I care about what happens after the notebook closes.
Hey, I'm Zayed :), an AI & Data Science student from Mumbai who's genuinely obsessed with building things that work, not just things that demo well.
I got into AI because I wanted to understand how machines actually learn not just call an API and call it a day. That curiosity pushed me toward ML pipelines, edge deployment, and more recently systems-level engineering in Rust and CUDA. Real constraints, real tradeoffs.
Right now I'm exploring the intersection of ML and low-level systems: writing fast data pipelines, understanding what happens under the hood, and building things that are measurably better not just "it works on my machine."
Building ML pipelines and tokenizers in Rust where Python can't go. Learning the hard way: no GIL, no runtime overhead, no hiding from the compiler.
Writing custom CUDA kernels for ML workloads. Understanding what actually happens on the GPU when a model trains not just calling .cuda().
MLflow experiment tracking, Docker-based deployments, and building systems that make models reproducible and production-ready not just accurate.
Actively looking to contribute to Rust ML crates like linfa and candle. Real codebases, real PRs not README fixes.
Implementing ML algorithms in Rust using the linfa ecosystem. Built a BPE tokenizer from scratch, then trained Decision Tree and Random Forest classifiers on Iris. Manual train/test splits, fixed random seeds, all without Python's safety net.
Production-ready Retrieval-Augmented Generation system enabling multi-document querying across PDFs with automatic source citations. Processes 27-page documents with sub-second response times and 90%+ answer relevance.
Fine-tuned GPT-2 on Shakespeare's sonnets to generate real-time poetry with preserved meter and rhyme schemes. Engineered custom loss functions optimizing for poetic structure.
Contributed to an ensemble CNN (Xception + InceptionV3) for road condition monitoring. Deployed on Raspberry Pi for edge inference, demonstrating low-cost AI solution for infrastructure monitoring. My contributions focused on model optimization and edge deployment.
Writing about what I actually build and break Rust for ML, CUDA internals, production data pipelines, and the gap between tutorials and real systems.
Follow along on @zayedansari2004 in the meantime.