My Projects
AgenticEvals (GSoC @ DeepMind)
Benchmark for evaluating LLMs across five foundational AI agent types (under mentorship from Google DeepMind in the Google Summer of Code program 2025).
Novel Attention Mechanisms (SNAP group @ SAIL)
Novel Attention Mechanisms for the largest open-source language model for sceince.
Semantic-Spatial Reasoning Benchmark (SAIL)
A benchmark to test semantic-spatial reasoning in MLLMs. Built using Gradio.
Stanford ACM Webpage
Maintaining the official Stanford ACM Webpage.
Resource Optimization for HDC ML models
Invented novel optimization methods for Hyperdimensional-Computing-based classifiers and generative models.
Contrastive Test-Time Scaling
Reasoning model that scales efficiently with test-time compute, beating state-of-the-art models.
Hyperdimensional Kolmogorov-Arnold Networks (HDKAN)
HDC-based KANs matching HDC-CNN vision accuracy but with improved parameter efficiency for edge devices.
Deep Behavioral Cloning with Synthetic MCTS Expert Data
Evaluated the viability of synthetic MCTS expert data for deep behavioral cloning in 2048.
E-Commerce Platform
Project Overview
A full-stack e-commerce platform built with React for the frontend and Node.js/Express for the backend. The application features user authentication, product catalog, shopping cart, payment processing, and order management.
Key Features
- User authentication and profile management
- Product catalog with search and filtering
- Shopping cart and checkout process
- Payment integration with Stripe
- Order tracking and history
- Admin dashboard for product management
Technologies Used
Image Recognition AI
Project Overview
An image recognition system built with TensorFlow and Python that can classify objects in images with high accuracy. The model was trained on a custom dataset and deployed as a web service.
Key Features
- Real-time image classification
- Support for multiple object detection
- Custom model training pipeline
- Web API for integration with other applications
- Performance optimization for edge devices
Technologies Used
Fitness Tracker App
Project Overview
A mobile application built with React Native that helps users track their fitness goals, workouts, and nutrition. The app includes features like workout planning, progress tracking, and social sharing.
Key Features
- Personalized workout plans
- Nutrition tracking and meal planning
- Progress visualization with charts
- Social sharing and community features
- Integration with fitness wearables
- Offline functionality
Technologies Used
Task Management Dashboard
Project Overview
A real-time task management dashboard for teams, built with Vue.js and Socket.io. The application allows teams to collaborate on projects, assign tasks, and track progress in real-time.
Key Features
- Real-time updates and notifications
- Kanban board for visual task management
- Team collaboration tools
- Time tracking and reporting
- Integration with calendar and email
- Customizable workflows
Technologies Used
LLM Agent Types Benchmark (GSoC @ DeepMind)
Project Overview
Designed, implemented, and tested the first benchmark for evaluating LLMs across all five foundational AI agent types (Google Summer of Code, mentored by DeepMind).