Aditi Bindal

and more..

OpenSorus: AI Maintainer Agent for GitHub Issues

OpenSorus is an autonomous GitHub Issues Maintainer Agent that augments open-source triage by aligning developer queries with semantically searched & indexed codebase contexts. Leveraging retrieval-augmented generation with LlamaIndex and Mistral’s Codestral model, the agent crafts and posts real-time, context-aware replies directly on GitHub.

Demo Interface is built with Gradio, and can be integrated seamlessly via GitHub App. This project has been recognized with the Mistral AI Choice Award at Hugging Face’s global Gradio Agents & MCP Hackathon.

Attention Is All You Need: Transformers Implementation From Scratch

A ground-up PyTorch implementation of the original Transformer architecture as proposed in the paper “Attention is All You Need”. This project dissects core components like multi-head self-attention, position-wise feedforward networks, and layer normalization, reconstructing them from first principles to offer deep transparency into the model’s mechanics.

The implementation is heavily inspired by Professor Dr. Arun Rajkumar’s lectures for the course Responsible & Safe AI Systems, and it serves both as an educational reference and a minimal baseline for understanding modern NLP architectures.

Dog Breed Classifier: Pretrained-CNN Image Classifier

A comparative image classification pipeline leveraging pre-trained convolutional neural networks (AlexNet, VGG, ResNet) to distinguish dog breeds with high accuracy across varied input scenarios. The system evaluates model robustness on both canonical pet datasets and real-world edge cases, enabling inference on custom user images through a lightweight batch script. Built to benchmark architectural performance in fine-grained visual recognition tasks.

This project was part of my hands-on lab work in the AI & ML Nanodegree Program, completed as a scholarship recipient of the AWS AI & ML Scholarship.

and more..