Kashif User

I am an ML Systems Engineer with four years of experience building production LLM systems and alignment workflows, training transformers from scratch, and developing low-level ML infrastructure in C++, Rust, and Python. I have designed execution runtimes for agent workflows, built hybrid knowledge‑base search over 200k+ documents, and led large‑scale transformer training and profiling experiments to improve throughput and inference behavior.

I have delivered end-to-end ML systems including fine-tuning domain LLMs (SFT/QLoRA), LLM agents that map inputs to SQL and generate business reports, and efficient inference engines in C++ with KV-cache and speculative decoding. My projects include pre-training a 360M LLaMA-style model, an autograd engine, a Rust n-d array library, and alignment work using Odds Ratio Preference Optimization.