Skip to content
View sarimahsan's full-sized avatar
💻
Focusing
💻
Focusing

Block or report sarimahsan

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
sarimahsan/README.md

Portfolio  LinkedIn  HuggingFace  Medium  ResearchGate  ORCID  Kaggle  Gmail



AI Research Engineer & Computer Engineering student at SSUET (GPA 3.5). Software Engineer at a UK firm, building production systems. I go low-level — designing transformer architectures from scratch, writing CUDA kernels, aligning LLMs with DPO & PPO, and probing hidden-state representations to forecast hallucinations before token generation.


141M-param transformer built from scratch · 15+ models on HuggingFace · 4 research papers · 8+ research projects



🔬 Research Focus

Large Language Models · Transformer Architectures · LLM Alignment (DPO, PPO, RLHF) · Mechanistic Interpretability · Efficient LLM Inference · Speculative Decoding · Knowledge Distillation · CUDA & GPU Programming



🛠️ Core Stack

AI / ML    Python PyTorch HuggingFace CUDA TensorFlow scikit-learn

LLM Systems    LangChain LangGraph vLLM OpenAI Gemini

Backend    FastAPI Node.js Django PostgreSQL MongoDB

Infra    Docker AWS GCP Linux GitHub Actions

Frontend    React Flutter JavaScript



📊 GitHub Stats

  


Pinned Loading

  1. arxiv-lens arxiv-lens Public

    AI-powered research tool that summarizes ArXiv papers, builds citation knowledge graphs, and lets you chat with any paper using RAG — built with FastAPI, HuggingFace, and React.

    Python 1

  2. gpt2-inference-engine-scratch gpt2-inference-engine-scratch Public

    A pure NumPy implementation of GPT-2 inference. No PyTorch in the forward pass — every operation (attention, layer norm, GELU, sampling) is written by hand using only numpy. Weights are loaded once…

    Jupyter Notebook

  3. Mini-Qwen_from_scratch Mini-Qwen_from_scratch Public

    A compact, educational implementation of a Qwen-inspired autoregressive language model built with PyTorch. The project includes transformer components, a custom AdamW optimizer, a Hugging Face data…

    Python

  4. nova nova Public

    A complete, self-contained implementation of a modern transformer language model built from scratch using PyTorch.

    Python

  5. interplens interplens Public

    InterpLens is an open-source mechanistic interpretability toolkit that accelerates neural network reverse-engineering with interactive circuit debugging and real-time telemetry. It bridges research…

    Python 2

  6. nova-moe nova-moe Public

    Nova-MoE is a clean, modular, from-scratch PyTorch implementation of a Mixture-of-Experts (MoE) Causal Language Model designed for controlled sparsity ablations against an active-parameter-matched …

    Python