Projects
GitHubA collection of ML, computer vision, and NLP projects.
GitHub Contributions
-
Multimodal Visual Question Answering with Generative AI utilizing LLM and Vision Language Models
Integrated Hugging Face pre-trained tokenizers, Visual Transformer for images, and LLMs for generating answers. Achieved 0.3 WUPS with RoBERTa and BEiT outperforming all 4 model combinations (ViT, DEiT with BERT, GPT). Deployed in Docker + Kubernetes.
-
Prompt Engineering and Few-Shot Learning with Flan-T5 for Dialogue Summarization
Performed prompt engineering on Flan-T5 using the DialogSum dataset with instructional prompts. Experimented with zero-shot and few-shot learning to assess impact on summary relevance and coherence.
-
Fine-Tuning Flan-T5 with PEFT (LoRA)
Fine-tuned Flan-T5 for dialogue summarization using Low-Rank Adaptation (LoRA), reducing computational cost while maintaining high ROUGE scores.
-
Retrieval Augmented Generation (RAG) Chatbot for 10Q Financial Reports
Retrieval Augmented Generation pipeline for querying SEC 10Q financial documents using LLMs. Built document ingestion, vector store retrieval, and conversational interface.
-
Transformer Architecture and Machine Translation from Scratch
Implemented Transformer model architecture from scratch. Trained on German→English translation (Multi30k dataset) with both greedy decoding and beam search.
-
Camera Calibration and Augmented Reality with Checkerboard
Applied calibration with checkerboard, camera pose estimation to align virtual objects within AR environment. Explored Harris corners and SURF feature detection, overlayed 3D virtual objects onto 2D video.
-
Vehicle Re-Identification using YOLOv8 and feature-based similarity
Feature-based similarity matching for vehicle re-identification across camera feeds using YOLOv8 detections and descriptor matching.
-
Image Colorization of Historical Paintings with GAN
Leveraged U-Net and pix2pix GAN to colorize grayscale historical paintings. Utilized CIELAB color space conversion and Patch discriminator for enhanced image-to-image translation.
-
Pothole Detection and Segmentation Jan 2020 – Jun 2020
Fine-tuned custom Mask R-CNN and YOLACT instance segmentation models for real-time pothole detection on Indian roads with PyTorch. Achieved 86% accuracy, 0.30 mAP on custom testing dataset.
-
Image Captioning using Encoder-Decoder Architecture Jan 2019 – Feb 2019
Image captioning application based on Neural Image Caption model using encoder-decoder architecture — pretrained CNN as encoder and LSTM as decoder.
-
Real-Time Twietter Social Media Sentiment Analysis Jul 2018 – Aug 2018
Real-time tweet sentiment feedback via time series graph, with past-week history and Watson API tone analysis on user-provided text.
-
Railway Ticketing System Built in Qt (C++ cross-platform framework) Apr 2018
Desktop app built in Qt (C++ cross-platform framework) for a railway ticketing system.
-
Location-Based Dengue Prediction using Gradient Boosted Trees Model on Weather Features Jan 2018 – Mar 2018
Android app with gradient boosted trees model on weather features, hosted on Django API. Real-time location-based dengue risk index. Showcased at Smart India Hackathon 2018 grand finale.
-
TV Script Generation and Face Generation using DCGAN Dec 2018 – Mar 2019
Generative models for realistic TV script generation and face synthesis using Deep Convolutional GAN (DCGAN).
-
Neural Style Transfer Dec 2018 – Mar 2019
Artistic style transfer using VGG-19 feature maps to blend content and style of images.
-
Sentiment Analysis using AWS SageMaker Dec 2018 – Mar 2019
Deployed a sentiment analysis model using AWS SageMaker with a full MLOps pipeline including training, evaluation, and endpoint deployment.
-
RL Navigation Agent (Banana Collection) Jul 2019 – Nov 2019
Trained a Deep Q-Network agent to navigate and collect bananas in a large square world using Unity ML-Agents.
-
Continuous Control — Double-Jointed Arm Agent Jul 2019 – Nov 2019
Trained a DDPG agent to control a double-jointed arm to reach target locations using continuous action spaces.
-
Collaborative Reinforcement Learning for Racket Sports Simulation Jul 2019 – Nov 2019
Trained two MADDPG agents to cooperatively play tennis, maximizing rally length across the net.
-
Local vs Global Temperature Trends Apr 2020 – Jun 2020
Analyzed local and global temperature data using SQL and compared trends with moving average visualizations.
-
A/B Test Analysis for E-Commerce Website Apr 2020 – Jun 2020
Statistical analysis of A/B test results to determine whether a new landing page improves conversion rates for an e-commerce site.
-
WeRateDogs Tweet Data Wrangling Apr 2020 – Jun 2020
Gathered, assessed, and cleaned WeRateDogs tweet data from multiple sources using programmatic and manual wrangling techniques.
-
Flight Delay and Cancellation Analysis Apr 2020 – Jun 2020
Explored flight cancellation and delay dataset with explanatory visualizations to communicate key findings.
-
Blog created in Django with PostgreSQL database Mar 2017
A blog developed in Django using PostgreSQL database.
-
Housing Price Prediction using Linear Regression, Lasso, Ridge, Boosting, and Random Forest Jan 2018
Predicted housing prices using linear regression, Lasso, Ridge, boosting, and random forest. XGBoost outperformed all other models.
-
Titanic Survivors Prediction using SVM, kNN, Logistic Regression, and Random Forest Oct 2017
Predicted Titanic survival using SVM, kNN, logistic regression, and random forest. Random forest gave the highest accuracy.