I'm Vaishnav, a Master of Computer Science student at Dalhousie University. I am a member of the MAPS LAB, focusing on Deep Reinforcement Learning techniques for Autonomous Navigation using maritime data. Recently, I have also been experimenting with meta-learning techniques for navigation.
Previously, I was a backend software engineer at Zopsmart.com and Ajackus.com. I have a total of 3 years of work experience experience, specializing in Golang and Python. I've developed a strong foundation in building scalable backend systems for e-commerce platforms.
MariNav is a data-driven maritime navigation environment that turns AIS vessel trajectories and ERA5 weather into a goal‑conditioned reinforcement learning testbed for smarter, safer routing across multiple origin–destination pairs on a hexagonal ocean grid. It assists by learning policies that choose direction and speed under dynamic winds, enforcing feasibility with action masking, and balancing fuel, time, and safety via shaped rewards informed by real traffic patterns. Built for reproducible research, MariNav evaluates PPO-based agents against graph baselines, showing that masking and domain-shaped rewards significantly improve stability and performance for practical decision support at sea.
Link to the research paper: https://arxiv.org/pdf/2509.01838
Link to python code for the research work: https://github.com/Vaishnav2804/MariNav
Youtube: https://www.youtube.com/watch?v=ovMWGQPOO_Q&t=1s
IEEE Digital Library: https://www.computer.org/csdl/video-library/video/2dOxPpG4xk4
This paper investigates Retrieval-Augmented Generation (RAG) for conversational AI chatbots to boost awareness of Indian government welfare schemes amid challenges like low literacy and linguistic diversity. It details scraping 2980 schemes from myScheme.gov.in, creating a multilingual pipeline with Gemini/Llama3 LLMs, Chroma DB, ASR/TTS for voice/text queries in regional languages. Experiments prove RAG outperforms plain LLMs in similarity, uncertainty, and hallucination metrics, favoring matched embeddings and larger models. We urge AI adoption for better scheme access, with open-source code on GitHub.
ISKON 1.pdfIn my senior year, while conversing with my professor after class, I learned that only 3% of Indians invested in equity markets due to a general risk aversion towards market volatility. This statistic was shocking, and I contacted numerous bank employees to understand better how Indians invest. Along with my professor and team, I aimed to create a solution, a simple website that interacts with AI/ML models that could help guide young Indians in making their first investments in equity markets. This work culminated in a research paper titled "A Hybrid Model for Stock Price Prediction using Machine Learning Techniques with CNN", which we published at an IEEE conference. Our work was awarded Best Paper at the 11th International Conference on Science and Innovative Engineering in 2021. This experience fascinated me with how neural networks work, especially LSTM and CNN.
Link to Research Paper: https://ieeexplore.ieee.org/abstract/document/9702382/