My Projects
A comprehensive showcase of my contributions to open source, professional work, and personal projects spanning machine learning, web development, and research.
Open Source Contributions
Contributing to the global developer community through meaningful code contributions and documentation improvements.
PDF Parser Preprocessing
AlibabaResearch
Contributed to creating a pipeline for the PDF parser 'VGT' using PDFPlumber, LayoutLMv3, BERT, and Tesseract to create `.pkl` files enabling smoother adoption of the provided library
MODNet Training Script
Image Matting
Provided the code to train the image segmentation model `MODNet` on private dataset since the code was 'private'.
Professional Projects
Production-grade projects developed during my tenure at various companies. These showcase my ability to deliver production-ready solutions at scale.
Digital Media Platform
Filmic Technologies
Architected and deployed a full-stack MVP connecting philanthropists and archivists. Built a comprehensive platform with multimedia uploads, real-time chat, role-based access control, and blochain-based payments.
Enabled the team at Filmic Technologies to present the product to existing archivists
Education Chatbot AI
Learno.AI
Spearheaded MVP development for K-12 education chatbot using RAG, and Streamlit for the frontend. Created a simple Python web-app with agentic LLMs and vector databases that worked as the required proof-of-concept.
Helped the co-founders to pitch their idea with the MVP to startup accelerator programs
PDF Document Parser
Learno.AI
Engineered advanced PDF parsing solution for visually complex documents by merging vision grid transformers with traditional parsers. Beat commercial-level pdf parsers on benchmark datasets like PubLayNet and DocLayNet
mAP of 0.82 @ IOU [0.50:0.95] on DocLayNet dataset
Image Matting & Enhancement System
Signimus
Fine-tuned an image segmentation model on a private dataset with 8.27% improvement in segmentation accuracy of that class.
8.27% improvement in segmentation accuracy
Photo editing app
Signimus
Created a photo-editing app that implemented multiple image-filters using GANs, Image Segmentation, and various other traditional image processing methods to achieve visual-rich and vivid results.
8.27% improvement in segmentation accuracy
Mentor Recommendation Engine
Expertrons
Coordinated development of comprehensive recommendation system using content-based and collaborative filtering to recommend 'mentors' to users based on their profiles, interests, and activity
Served 250K active users, 90% API performance improvement
Personal Projects
Side projects and research work that showcase my passion for learning and experimenting with new technologies.
F1 Winner Prediction
A machine learning project that predicts Formula 1 race winners using historical race and driver data. This repository contains data preprocessing pipelines and notebooks implementing various classification models to compare performance and identify the best predictor.
New Orleans Police Respone Time Prediction
A machine learning project to predict police response times for the New Orleans Police Department using incident data from 2025. This project analyzes emergency response patterns to help optimize resource allocation and improve public safety services.
Custom Cover Letter Generator
Customize cover letter and resume to the job description using LLMs with pre-defined templates. Uses langchain, pydantic, and FastAPI.
CNNs, RNNs, and LSTM architectures
A deep learning project implementing VGG-16 and ResNet-18 architectures for image classification on a 30K multi-class image dataset. This repository contains comprehensive experimentation with different optimizers and regularization techniques to achieve target accuracies above 75%. The project also includes time-series forecasting using RNNs/LSTMs and sentiment analysis implementations to compare model performance across various neural network architectures.
Autoencoders and Transformers
An advanced neural network project focused on autoencoder-based anomaly detection and transformer models for natural language processing tasks. This repository contains custom autoencoder implementations for time-series anomaly detection achieving over 80% accuracy, along with complete transformer architectures built from scratch using multi-head attention mechanisms. The project includes fine-tuned BART models for text summarization with comprehensive evaluation using ROUGE, BLEU, and BERTScore metrics.
Legal Document Analyzer (AI4Good)
A legal document analysis project built as a FastAPI application using LLMs to streamline contract review and risk assessment. This repository provides endpoints for uploading PDFs, DOCX, or text files, generating concise summaries, rating document complexity on a 1-10 scale, detecting red-flag clauses, extracting financial figures and deadlines, and identifying vague or ambiguous terms. Built as part of the AI4Good hackathon.
Technologies & Skills
Specialized tools and technologies organized by domain expertise