Building Intelligent Systems with Passion & Precision

ROUV Software Suite

Developed a complete software stack for a Remotely Operated Underwater Vehicle, including the user interface, control algorithms, and hardware interfaces. This software helped junior team to secure 2nd spot in AMUROVC3.0, a national level underwater competition.

AI-Powered Text Summarizer

This project is a text summarization/keypoint extraction tool that uses a Python backend powered by Google's Gemini Flash to generate summaries and key points from text. It also includes a Chrome extension that allows you to summarize web pages or selected text directly in your browser.

Hybrid Retrieval

Local hybrid PDF retrieval system for searching through a collection of PDF documents. It extracts text from PDFs, generates embeddings, and uses a hybrid search approach to find relevant documents based on a user's query.

Singapore AUV Competition 2022

Contributed to the computer team by testing the vision system of the AUV and developing a custom dataset for various vision tasks.

Differential ViT

Implemented and adapted Differential Transformer paper by Ye et al. to create Differential ViT.

RLDM UAV

Reinforcement Learning-driven integrated Detection and Mitigation of UAV GPS spoofing attacks, an end-to-end framework that relies on GPS data and onboard camera inputs, eliminating the need for additional sensors. Utilizes a Deep Q-Network (DQN) to facilitate dynamic GPS spoofing detection and adaptive switching between GPS-based and vision-based navigation policies based on GPS signal reliability.

Implemented a U-Net architecture for semantic segmentation of cell nuclei in medical images, trained on the 2018 Data Science Bowl dataset.

Neural Network from Scratch

Created a shallow neural network using only Numpy to identify handwritten digits, trained on the MNIST dataset.

ESM-3 for DARPins

Fine-tuned the ESM-3 protein language model on a curated dataset of Designed Ankyrin Repeat Proteins (DARPins). Focused on fine-tuning ESM-3 to enhance its capabilities for generating stable and high-affinity DARPin sequences.

Multiagent Reinforcement Learning Aerial Dogfight

Developed a scalable multiagent reinforcement learning based aerial dogfighting simulation. Trained AI agents to perform complex aerial combat maneuvers and achieve higher-level objectives in dogfighting scenarios.