It is a LoRA adapter for Microsoft's phi-2 that helps in context-aware conversations, trained on 50K+ conversations of HuggingFace's HyperThink dataset. It is available on HuggingFace (getting over 100+ downloads).

It is a web app that predicts nifty 50 stock prices using LSTM algorithm. It is an upgraded version of my previous Random Forest based model. The code is available on GitHub and the demo is available on HuggingFace.

It is a web app that detects anomalous maritime vessel behavior from AIS (Automatic Identification System) data using autoencoders. Checkout the model on HuggingFace.

It is a transformer based decoder built from scratch. After learning the basics of attention is all you need research paper, I built this decoder to understand the inner workings of transformers. The code is available on GitHub.

It is an AI-powered animation generation platform that transforms natural language prompts into mathematical animations using the Manim library. The backend is a production-ready FastAPI service that orchestrates a complex pipeline of prompt expansion, code generation, and video rendering.

It is a part of a RAG-Document Summarizer project. This pipeline ingests PDF/txt files, extracts the text, then embeds it using Pathway's Framework and uses its llm to generate a summary or to answer questions.

It is a web app that helps in analyzing and visualizing any CSV or Excel file. It is built using Streamlit and Python. The code is available on GitHub.

It was an internship project under Academy of Skill Development, where our team analyzed the Nifty 50 dataset of 2023-2024 and provided insights on the market trends, volatility patterns, and statistical insights to empower traders, analysts, and enthusiasts with actionable intelligence. The report is also available in the GitHub repository.