Skip to content
View sumony2j's full-sized avatar
β›³
Learner
β›³
Learner

Block or report sumony2j

Report abuse

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

Report abuse
sumony2j/README.md


πŸ‘‹ Hi there, I'm Sumon Singh

Welcome to my GitHub profile!


πŸ”­ I’m currently working on

Distributed AI/ML systems, infrastructure tooling, and performance optimization in large-scale training environments.

🌱 I’m currently learning

Horovod, LLM training/inference pipelines, GPU Direct RDMA, and exploring internals of NCCL, UCX, and GDRCopy.

πŸ‘― I’m looking to collaborate on

Open-source projects in AI infrastructure, systems programming, and performance engineering.

πŸ’¬ Ask me about

Python, C/C++, Linux, Docker, Jenkins, Bash, Java, Git, AWS EC2, Automation, Distributed Training, and Open-source contributions.

⚑ Fun fact

My commit messages tell a story – a tragic tale of bugs squashed, features conquered, and the occasional coffee spill. It's a novel in progress.


πŸš€ Highlights of My Work

🧠 AI/ML & Distributed Systems

  • Fine-tuned LLMs and optimized inference pipelines for performance and scalability.
  • Built and containerized Horovod-based distributed training setups using Docker.
  • Debugged NCCL for improved communication efficiency in multi-node setups.
  • Integrated UCX in deep learning environments.
  • Applied GPU Direct RDMA and GDRCopy to accelerate memory transfer.
  • Benchmarked using nccl-tests and ib_perf.

βš™οΈ DevOps, Automation & Infrastructure

  • Built CI/CD pipelines with Jenkins, Git, AWS EC2, and Python.
  • Developed CLI tools in C and Python.
  • Wrote automation scripts in Bash and Python for deployments and monitoring.
  • Built packet manipulation tools and custom JSON parsers.

πŸ“š Engineering & Open Source

  • Integrated data algorithms using JNI, shared libraries, C/C++.
  • Contributed documentation and internal tooling.
  • Advocates for clean, performance-first code and OSS collaboration.

πŸ† Top Projects – Live Demos & Source Code

🧠 SeedGPT-Basic

A minimal decoder-only LLM from scratch

Tech: PyTorch, Gradio

GitHub Live Demo

🧠 SeedGPT-StoryCrafter

AI for stories, tales & imagination

Tech: PyTorch, Hugging Face, Streamlit

GitHub Live Demo

🧠 SeedGPT-22M

Compact, Efficient Small Language Model

Tech: PyTorch, Hugging Face, Streamlit

GitHub Live Demo

🩺 DeepCareX

DeepCareX is an AI-powered healthcare system leveraging machine learning models for intelligent health insights

Tech: TensorFlow, Keras, Flask, Docker

GitHub Live Demo

🎨 ArtVenture

Real-time AI filters and image transformations.

Tech: OpenCV, Streamlit, Docker

GitHub Live Demo

πŸ“« Connect With Me

LinkedIn Gmail

πŸ’» Skilled In

πŸ–₯️ Operating Systems

Windows Ubuntu

πŸ”€ Languages

Python Shell Script Java C C++

🧠 ML/DL

TensorFlow PyTorch Keras DeepSpeed LLM NCCL

🌐 Web & DB

Flask HTML CSS Bootstrap SQLite

βš™οΈ DevOps Tools

Docker Jenkins AWS EC2 Git Vagrant



πŸ“ Publications

  1. πŸ” Unveiling the Veil: A Comprehensive Assessment of Privacy and Security in Amazon Alexa
    International Journal of Innovative Science and Research Technology

  2. 🧘 Yoga pose classification from images using transfer learning
    International Journal of Innovative Research in Technology

ReadMe Card


🐳 Docker Images

πŸ“¦ Find my Docker images on DockerHub: sumon2j

🩺 DeepCareX

Multi-disease diagnostics AI system.

docker pull sumon2j/deepcarex:latest

DockerHub DeepCareX

🎨 ArtVenture

Image AI filters and transformations.

docker pull sumon2j/artventure

DockerHub ArtVenture

πŸ† GitHub Trophies


πŸ“Š Stats

Sumon's GitHub stats Top Langs


Pinned Loading

  1. SeedGPT-22M SeedGPT-22M Public

    SeedGPT is a lightweight, 22M-parameter Transformer LLM for efficient text generation and chat, with Streamlit-based real-time UI.

    Python

  2. DeepCareX DeepCareX Public

    DeepCareX is an AI-powered healthcare system leveraging machine learning models for intelligent health insights

    Dockerfile 7 4

  3. SeedGPT SeedGPT Public

    SeedGPT is a minimal decoder-only LLM from scratch, inspired by andrej karpathy's tutorials, generating text token-by-token based on input word prompts.

    Python 1

  4. ArtVenture ArtVenture Public

    Apply various filters on images

    Python 1

  5. Transfer-Learning Transfer-Learning Public

    Yoga pose classification from images using transfer learning approach.In this project, a total of 1551 images representing 5 distinct yoga postures were used. Transfer learning was employed, utiliz…

    Jupyter Notebook

  6. Simple-BPE-Tokenizer Simple-BPE-Tokenizer Public

    A pure Python implementation of Byte Pair Encoding (BPE) tokenizer. Train on any text, encode/decode with saved models, and explore BPE tokenization fundamentals.

    Python