Skip to content
View couder-04's full-sized avatar
8 followers · 33 following

Block or report couder-04

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
couder-04/README.md 
██████╗  █████╗ ██████╗ ████████╗██╗  ██╗    ████████╗ ██████╗ ███╗   ███╗ █████╗ ██████╗
██╔══██╗██╔══██╗██╔══██╗╚══██╔══╝██║  ██║    ╚══██╔══╝██╔═══██╗████╗ ████║██╔══██╗██╔══██╗
██████╔╝███████║██████╔╝   ██║   ███████║       ██║   ██║   ██║██╔████╔██║███████║██████╔╝
██╔═══╝ ██╔══██║██╔══██╗   ██║   ██╔══██║       ██║   ██║   ██║██║╚██╔╝██║██╔══██║██╔══██╗
██║     ██║  ██║██║  ██║   ██║   ██║  ██║       ██║   ╚██████╔╝██║ ╚═╝ ██║██║  ██║██║  ██║
╚═╝     ╚═╝  ╚═╝╚═╝  ╚═╝   ╚═╝   ╚═╝  ╚═╝       ╚═╝    ╚═════╝ ╚═╝     ╚═╝╚═╝  ╚═╝╚═╝  ╚═╝

Hi, I'm Parth Tomar 👋

Aspiring ML Engineer & Builder · IIT Patna · New Delhi, India

LinkedIn Codeforces Gmail Profile Views

🧠 About Me

parth = {
    "institute"   : "IIT Patna — Computer Science & Engineering",
    "focus"       : ["Machine Learning", "Deep Learning", "NLP", "Computer Vision"],
    "currently"   : "Building multimodal AI systems & exploring CV architectures",
    "goal"        : "Ship intelligent systems that solve real problems",
    "fun_fact"    : "I implemented backprop from scratch in NumPy before touching PyTorch"
}

🏗️ Featured Projects

🔬 VisionBench

Benchmarking 6 architectures on Cat vs Dog classification

Systematic comparison of Logistic Regression, SVM, Custom CNN, ResNet-34, EfficientNet-B0, and ViT-Base — from pixel vectors to attention patches.

PyTorch ViT ResNet EfficientNet Benchmarking

⭐ What I learned: ViT needs far more data than CNNs at this scale

🧮 NeuroSketch

MNIST recognition built from scratch — zero deep learning frameworks

Full training pipeline, experiment tracker, dashboard, and interactive canvas app using only NumPy. Backprop by hand.

NumPy Neural Networks From Scratch Interactive

⭐ The project that taught me what autodiff actually does

⚡ CortexCPP

A deep learning framework written in C++

Custom tensor runtime, dynamic NN & CNN construction, shape inference, architecture validation, and forward/backward pass — all in C++.

C++ Tensor Runtime Deep Learning Systems

⭐ Because understanding frameworks means building one

🤖 RepoRelic

AI-powered developer intelligence platform

Deeply analyzes GitHub repositories via a multi-stage autonomous pipeline — architecture mapping, code intelligence, and insights.

Python LLM Agentic AI Code Analysis

⭐ Think of it as a senior engineer reviewing any repo in seconds

🎬 TranscriptMind

NLP-powered YouTube transcript summarization

Transforms long-form transcripts into context-aware summaries using TextRank, BART, PEGASUS, and T5 — with model comparisons.

NLP BART PEGASUS T5 TextRank

⭐ 4 summarization models, 1 clear winner (WORKING)

👁️ CortexVision

Real-time multimodal scene understanding

Detection + depth estimation + segmentation + tracking + VLM fusion — all in one pipeline for real-time environment comprehension.

Computer Vision Multimodal VLM Real-time

⭐ The most complex system I am building so far

🛠️ Tech Stack

Languages

Python C++ JavaScript

ML / DL

PyTorch TensorFlow scikit-learn NumPy OpenCV HuggingFace

Tools & Infra

Git Jupyter Linux Docker

📊 GitHub Stats

🏆 GitHub Trophies

trophy

📈 Contribution Activity

Parth's Activity Graph

"The best way to understand a system is to build it from scratch."

Let's connect and build something that matters.

LinkedIn

Pinned Loading

  1. VisionBench VisionBench Public

    Benchmarking Cat vs. Dog classification across 6 architectures — Logistic Regression, SVM, Custom CNN, ResNet-34, EfficientNet-B0, and ViT-Base. From pixel vectors to attention patches.

    Jupyter Notebook 1

  2. SceneSenseAI/CortexVision SceneSenseAI/CortexVision Public

    Persistent world modeling from monocular video — depth fusion, metric scene graphs, object memory, and predictive reasoning

    Jupyter Notebook

  3. NeuroSketch NeuroSketch Public

    End-to-end MNIST digit recognition system built from scratch with NumPy, featuring training pipeline, experiment tracking, dashboard, and interactive canvas app.

    Jupyter Notebook 2

  4. RepoRelic RepoRelic Public

    an AI-powered developer intelligence platform that deeply analyzes repositories through a multi-stage autonomous pipeline

    Python 2

  5. TranscriptMind TranscriptMind Public

    TranscriptMind is an NLP-powered transcript summarization system that transforms long-form YouTube transcripts into concise, context-aware summaries using TextRank, BART, PEGASUS, and T5.

    Jupyter Notebook 1

  6. CortexCPP CortexCPP Public

    A modular deep learning framework in C++ featuring a custom tensor runtime, dynamic neural network and CNN architecture construction, shape inference, architecture validation, model analysis, forwa…

    C++ 1