Skip to content
View SahilSharma1306's full-sized avatar
  • Jadavpur University
  • Kolkata, India

Block or report SahilSharma1306

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
SahilSharma1306/README.md

Typing SVG


Profile Views

🧠 About Me

Name:       Sahil Sharma
University: Jadavpur University, Kolkata
Degree:     B.E. Instrumentation & Electronics Engineering (Class of 2027)
Focus:      AI/ML Engineering  ×  Hardware Design
Philosophy: "If I can't build it from scratch, I don't understand it."

I operate at the intersection of intelligent software and bare-metal hardware. From training a 148M-parameter Transformer on raw tensors to designing reconfigurable systolic arrays in Verilog — I build systems from the ground up, because understanding how something works is the only way to make it work better.


🚀 Featured Projects

Python · PyTorch

A 148M-parameter Encoder-Decoder Transformer built entirely from raw tensors. No nn.Transformer. No shortcuts.

  • ✦ RoPE positional encoding
  • ✦ RMSNorm normalization
  • ✦ SwiGLU activation
  • ✦ Multi-GPU DDP training
  • ✦ Mixed-precision (AMP) pipeline
  • ✦ Trained on 1.1M EN→FR sentence pairs

Verilog · Hardware Design

A reconfigurable systolic array architecture supporting both dense and sparse computation modes.

  • ✦ Configurable dense/sparse modes
  • ✦ Designed at the RTL level
  • ✦ Optimized for matrix acceleration
  • ✦ Built for AI/ML workloads
  • ✦ Clean, synthesizable Verilog

Python · scikit-learn · EEG

AI-powered ADHD screening system using raw EEG signals. IBM Z Datathon submission.

  • 99.91% classification accuracy
  • ✦ EEG signal processing pipeline
  • ✦ Feature engineering from brainwave data
  • ✦ End-to-end ML pipeline on IBM Z

8051 Assembly · Hardware Interfacing

Interrupt-driven PWM waveform generator with live UART serial control.

  • ✦ 100 Hz Software PWM via Timer 0
  • ✦ 8-bit DAC Bipolar Output (±2V)
  • ✦ Real-time UART control (9600 baud)
  • ✦ Bare-metal 8051 assembly

🔜 What's Next

Always building...

Currently exploring:

  • ✦ Custom RISC-V accelerators for inference
  • ✦ Efficient attention mechanisms
  • ✦ Hardware-software co-design for edge AI
  • ✦ Quantization & model compression

🛠️ Tech Stack

Languages & Frameworks

Python PyTorch Verilog Java MATLAB

ML & Data Science

NumPy Pandas scikit-learn HuggingFace TensorBoard

Tools & Platforms

Git Linux VS Code Kaggle


📊 GitHub Stats

GitHub Stats GitHub Streak
Top Languages

🐍 Contribution Graph

github-snake

🤝 Let's Connect

LinkedIn GitHub Gmail


Quote



⚡ Built from scratch — just like everything else I make.

Pinned Loading

  1. transformer-from-scratch transformer-from-scratch Public

    A pure-PyTorch Transformer built from raw tensors — RoPE, RMSNorm, SwiGLU, DDP, AMP — trained on 1.1M EN→FR sentence pairs.

    Python

  2. versatile-systolic-array versatile-systolic-array Public

    Verilog RTL implementation of the VERSA reconfigurable systolic array with dense and sparse computation modes

    Verilog

  3. IBMZDatathon IBMZDatathon Public

    ADHD affects focus, impulse control, and emotions, impacting 5–7% of children and 5% of adults. Diagnosis is often delayed and inconsistent due to reliance on self-reports and clinical interviews.O…

    Jupyter Notebook