I build ML systems at the compiler–model boundary.
- GSoC 2024 contributor to
rustc(sandboxed deterministic proc macros via Wasm) - ML intern at IIT Hyderabad (ML systems, evaluation rigor, synthetic-data workflows)
- Compiler/runtime systems for ML workloads
- Reproducible model training + evaluation pipelines
- Applied LLM tooling and experimentation in JAX/Rust
- Added experimental support to run procedural macros in WebAssembly
- Worked on compiler-runtime integration and token-stream communication
- Project Link
- Collection of models in pure JAX (language + vision experiments)
- Focused on clean training loops and reproducibility
- Repository
- Browser-like app using Typst documents instead of HTML
- Rust core + SwiftUI frontend
- Repository
- Minimal JAX/NNX training scaffold with multiple attention backends
- Includes checkpointing and data pipeline support
- Repository
I’m currently working on ML systems + compiler workflows, especially:
- evaluation correctness
- data quality
- reproducible training loops
- Email: apurva.jpr@gmail.com
- CV: link
- X: @mav3ri3k