ML/Software Engineer. 4+ yrs enterprise distributed systems (Deloitte, 23M+ scale), MSc Advanced Computer Science (AI) @ Leeds.
Open Source & Core Contributions:
- PyTorch Core: Contributor to upstream PyTorch (
torch.nn.functional.nll_lossMPS forward migration on Apple Silicon, merged intopytorch:main). - SFHQ-VirtualID: Synthetic identity-conditioned benchmark for machine unlearning (750 identity clusters, 75k portraits, CC0 seeds, Slurm HPC array, published with DOI).
serverless-github-calendar: Zero-runtime-dependency GitHub contribution heatmap via GitHub Actions.msc-project-core-implementation: Dual-head ResNet-18 evaluation framework for identity-level unlearning baselines.
Building:
- Breadwinner: Local-first, zero-hallucination career engineering studio & AST compiler in Textual. Deterministic verification, offline Ollama/LM Studio support, and headless Chromium ATS PDF generation.
- MLX Ternary: High-performance Apple Silicon runtime for 1.58-bit ternary language models with hardware vector dot product operations (
dot(half4, half4)) and Split-K parallelization. - GraphContext MCP: Structural AST and dependency graph engine for AI coding assistants (Proprietary commercial product).
1,100+ contributions in the last year. Writing & systems essays: faizpalwala.tech
