- Fixed LLVM codegen bug by mapping SIMD comparisons to FCMP UNE, correcting vector NaN evaluation
- Wrote tests validating IR emission and SIMD lane boolean reduction behavior
Education & Skills
Northwestern University
September 2026 – June 2030 (Anticipated)
Computer Science & Operations Research (IEMS)
Evanston, IL
NU Formula Racing (FSAE): Firmware engineer; using C to build high voltage battery management system.
Languages
Python
C++
Mojo
Rust
Go
JavaScript
Tooling & Frameworks
PyTorch
LLVM
Cranelift
Google OR-Tools
Jax/Flax
LightGBM
Optuna
Docker
Knowledge/Domains
Multi-Threading
Optimizations
Clustering Algorithms
ML
Git
Linux
SIMD
C-FFI
Projects & Open Source Contributions
- Built zero-copy interop between Mojo and the Rust Polars dataframe library via Apache Arrow C Data Interface
- Implemented an RAII wrapper to synchronize Mojo
__deinit__with Arrow C release callbacks to prevent leaks. - Implemented SIMD reduction kernels (sum, mean, std) via Mojo’s vectorize primitives.
pixi add polars-mojo
- Developing a pure Mojo machine learning library with Sklearn ergonomics and zero Python runtime dependencies.
- Leveraged strict traits, static polymorphism, and generic SIMD types across float/int bit-widths to optimize ML models (Linear models, Random Forests, K-Means). Average of 1.9 times faster than Sklearn. See benchmarks
- Applied LAPACK/BLAS to handle linear algebra computation.
pixi add strata
- Built an end-to-end compiled language called Salmon, that targets native machine code via LLVM IR
- Implemented SSA code-gen using LLVM C++ API with scope management & dynamic list runtime structs.
- Designed deterministic resource cleanup via a LIFO defer unwinding stack.
- A Python library for clustering categorical data using K-Modes with various distance metrics (3000+ downloads).
- Implemented multi-threading, GPU acceleration, and JIT optimizations for better performance.
pip install kluster-fudge
- Contributed upstream bug fix to CubeCL, a multi-platform GPU compute language extension for Rust.
- Resolved critical packaging automation failure by debugging and patching CI/CD scripts for the Mojo toolchain.
- Built Rust library for survival analysis and a library for K-Modes algorithm (combined 2000 downloads).
- Implemented Newton-Raphson and AdamW optimizers, with L1/L2 regularization for feature selection.
Work Experience
Research Intern
MIT Jameel Clinic • Cambridge, MA
October 2024 – Present
- Performed clustering analysis on public health data; co-authoring 2 in-progress papers targeting PLOS One.
- Developed novel method to find the best clustering from clusterings generated with different hyperparameters.
- Replicated & verified cancer prediction computational models in published papers.
- Co-designed a custom featurizer for a cancer prognostic machine learning model with an MIT PostDoc.
- Had several of my projects featured in an MIT blog post: See Blog Post