ML Research Engineer · scientific computing, neural operators, and applied machine learning
ML Research Engineer at Aeris UK, based in London. MSc Data Science (Distinction). I work at the intersection of numerical methods and machine learning — reinforcement learning and simulation software, PDE solvers and neural-operator surrogates, distributed training pipelines, and LLM-driven agents. I care about code that is correct, tested, and honest about its trade-offs: most of my recent projects ship with a validation suite, CI, and benchmarks rather than just a notebook.
- Currently building reinforcement-learning and simulation software at Aeris UK.
- Research interest in operator learning for PDEs and high-performance numerical computing.
- Comfortable across the stack: C++/OpenMP kernels, PyTorch (DDP/FSDP), and full-stack deployment.
- 15+ ML models deployed across research and production settings.
- Reach me at abdullahrasheed45@gmail.com or on LinkedIn.
| Repo | Merged pull request(s) |
|---|---|
| huggingface/transformers (165k⭐) | #47663 fix: correct text input validation logic in 8 multimodal processors (and → or) |
| EleutherAI/lm-evaluation-harness (14k⭐) | #3971 fix: prevent ValueError when batch_size="auto" is passed to neuronx model · #3970 fix: prevent ValueError when batch_size="auto:N" is passed to API models |
Also have 16 pull requests currently under review across 12 other repositories:
Open pull requests
| Repo | Pull request(s) |
|---|---|
| stanfordnlp/dspy (38k⭐) | #10112 fix: deduplicate signatures by content in get_dspy_source_code (fixes silent drop of dynamic signatures) · #10085 Roll nested track_usage() scopes up into the enclosing tracker |
| Lightning-AI/pytorch-lightning (31k⭐) | #21866 Bump rich upper bound to <15.1.0 to support rich 15.0 |
| pyg-team/pytorch_geometric (24k⭐) | #10762 validate(): raise on numpy arrays stored as tensor attributes · #10756 Verify TLS certificates when downloading datasets |
| huggingface/datasets (22k⭐) | #8368 Reject duplicate column names in Dataset.select_columns |
| huggingface/peft (22k⭐) | #3477 Remove PEFT forward hooks when the wrapped block raises |
| EleutherAI/lm-evaluation-harness (14k⭐) | #3951 Include generation_kwargs in the request cache key |
| hydra-ecosystem/hydra (11k⭐) | #3333 fix(optuna-sweeper): pass values=None when marking trial as FAIL, add NaN check |
| huggingface/accelerate (9.8k⭐) | #4138 Reject sharded-checkpoint shard paths that escape the checkpoint folder |
| bentoml/BentoML (8.8k⭐) | #5676 Keep every URL value of a multipart list field |
| cvxpy/cvxpy (6.3k⭐) | #3470 Stop internal parameter checks warning about sparse .value reads |
| huggingface/lighteval (2.5k⭐) | #1314 Fix always-true conditional that made DEFAULT_FORMAT unreachable · #1313 Fix litellm judge sending max_tokens as a 1-tuple |
| Lightning-AI/torchmetrics (2.5k⭐) | #3441 fix(validation): correct argument validation in auroc, logauc, and pesq metrics · #3440 fix(validation): correct 'and' to 'or' in argument validation across multiple metrics |
| Project | What it is | Stack |
|---|---|---|
| fno-darcy-flow | Fourier Neural Operator surrogate for 2D Darcy flow, with a from-scratch finite-volume data generator and 1-vs-N-GPU scaling benchmarks. Trains identically under DDP and FSDP/ZeRO-3. | PyTorch, DDP/FSDP, NumPy |
| darcy-cpp | Matrix-free multigrid finite-volume solver for 2D Darcy flow. Grid-independent CG iteration count, OpenMP-parallel, with zero-copy Python bindings — up to 16× faster than a sparse direct solve. | C++17, OpenMP, pybind11 |
| Adaptive-Micro-Gesture-Recognition | Master's thesis: a 3D CNN that recognises subtle hand movements as computer commands to improve accessibility for motor-impaired users (94.6% accuracy). | TensorFlow, OpenCV, Flask |
| Grammify | A T5 Seq2Seq grammar-correction app with real-time error analysis, deployed on Hugging Face Spaces. | Transformers, FastAPI, Streamlit |
| crypto_trading_ai_agents | Agentic system that forecasts price movements and produces buy/sell signals. | Python, LLM agents |
Languages · Python, C++, SQL, JavaScript/TypeScript, R
ML / scientific ·
Systems / tooling ·
Web ·
- ML Research Engineer — Aeris UK (current) · reinforcement learning and simulation software
- Data Analyst — Digital Pulse 360
- Full-Stack Developer — ATnR, Karachi
- MSc Data Science — Anglia Ruskin University, Cambridge (Distinction)
- BSc Economics — Institute of Business Administration (IBA), Karachi (Dean's Honour List)
Certifications · Google Advanced Data Analytics · Google AI Essentials · Google Data Analytics Professional · Tableau Specialization (UC Davis)