8000 amartya116 (Amartya Dudhe) Β· GitHub
[go: up one dir, main page]

Skip to content
View amartya116's full-sized avatar
πŸ˜‰
πŸ˜‰
  • Germany
  • 13:46 (UTC -12:00)

Block or report amartya116

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.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. 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
amartya116/README.md

Amartya Dudhe

M.Sc. Computer Science, University of Marburg. I work on AI engineering and systems-level C++, and I learn things by building them from the bottom up β€” most recently a deep learning library with my own SIMD kernels and autograd engine, because calling a framework and understanding one are different skills.

Previously: agentic AI systems and MCP servers at Panta Index (Giessen); full-stack MERN development at Thinkonic Softwares (Nagpur).


Projects

MiniPyTorch β€” a PyTorch-style deep learning library in C++20 with a Python interface via pybind11. Tensor operations, a reverse-mode autograd engine, AVX2 SIMD kernels for element-wise and reduction operations, and a CMake/vcpkg build. Blocked GEMM, multithreading and a CUDA backend are in progress.

Dataset augmentation with LLMs (university project, ongoing) β€” LLM-based techniques for synthetically expanding training datasets, with ChromaDB for vector storage and retrieval and DuckDB for analytical queries. The focus is on measuring whether augmented data actually improves downstream performance.

MCP servers and RAG pipelines β€” Model Context Protocol servers giving LLM agents structured access to data through reusable tool interfaces; retrieval pipelines over unstructured sources.


Technical background

Languages β€” C++20, Python, JavaScript, TypeScript, PHP, SQL

Systems β€” AVX2 SIMD vectorisation, pybind11, CMake, memory management, multithreading, performance profiling

AI engineering β€” agentic systems, LLM integration, MCP servers, RAG and vector search (ChromaDB), embeddings, evaluation design

Web β€” MongoDB, Express, React, Node.js, REST APIs

Tools β€” Git, Linux, Docker, DuckDB


How I work

I believe in Feynman's line: what I cannot create, I do not understand. That is why I build things from scratch.

When I'm working I speak less β€” technical and precise, only what is needed to convey the point.

I like studying things at a fundamental level, as deep as I can go. I enjoy hard problems, and I work well in a team.


Currently learning

Writing CUDA kernels, C++ concurrency, and distributed systems. Longer term I'm interested in robotics, embedded systems, and applying computing to problems in the physical sciences.


amartyadudhe13@gmail.com

Popular repositories Loading

  1. minipytorch minipytorch Public

    A custom made pytorch like library from scratch. AI only used to correct and suggest edge cases, rest all written by me.Written in Python and C++.

    C++ 1

  2. magnetron magnetron Public

    Forked from MarioSieg/magnetron

    A zero-dependency ML framework in C with a modern Python API for full control over execution and memory.

    C 1

  3. x-ray-pneumonia-lung-detection x-ray-pneumonia-lung-detection Public

  4. face-mask-detector face-mask-detector Public

    Python

  5. automated-Hydroponics-system automated-Hydroponics-system Public

  6. A3CV A3CV Public template

    Forked from amanpache14/A3CV

    Applications Of Deep Learning using computer vision

    HTML

0