Projects | Hari Veeramallu

Computer Vision

DiffTransBEV - Improved BEV using Diffusion models

Developed an innovative and novel deep learning architecture integrating SwinV2 LSS, DPM and Scalable Diffusion Transformers to generate accurate BEV representation in Autonomous Vehicles from 6 RGB camera sensor inputs.

Evaluating Diver Detection - YOLOv8 vs. Transformer Models

Demonstrated the superior performance and computational efficiency of YoLOv8 over DETR architecture on the Video Diver Dataset (VDD).

Primate Pose Estimation

Designed a deep learning model using rCNNs and CPM for the OpenMonkeyChallenge to estimate monkey poses in natural habitats. It tracks 17 monkey pose landmarks, with its accuracy measured by MPJPE. The model achieved an MPJPE of 0.217, enhancing wildlife monitoring and behavioral studies through AI.

Autonomous Vehicles

Motion Prediction for Autonomous Vehicles

Developed an advanced deep learning model to accurately predict the trajectories of surrounding vehicles in relation to self-driving vehicles. This model integrates Polyline-LSTM-MLP architecture with transformer-based encoders, effectively analyzing both the movement history of nearby agents and the intricate patterns of road networks.

Self-Driving Vehicle Control

Crafted a sophisticated vehicle controller for autonomous navigation through predefined waypoints in the CARLA simulation environment. This controller combines Proportional-Integral-Derivative (PID) methodology for precise longitudinal control with the Stanley controller for lateral management, ensuring smooth and accurate maneuvering of the autonomous vehicle.

Natural Language Processing

Stock Performance Forecasting using Meeting Transcripts

Built an NLP pipeline to predict stock performance from stakeholder meeting transcripts. It summarizes transcripts with Huggingface T5/GPT LLMs and analyzes sentiment with Huggingface BERT LLMs, offering insights into market movements based on stakeholder communications.

Knowledge Graph Augmented Common-sense Reasoning

Developed a deep learning model with a Knowledge Graph for answering commonsense multiple-choice questions. It integrates GCN, LSTM, and HPA, achieving 40% accuracy, showcasing promise in AI-driven commonsense reasoning as compared to human benchmarks.

Data Analysis

OList E-Commerce Analysis

Developed a data warehouse for Brazilian eCommerce company OList, covering over 96,000 customer sales. The project included RFM and Spatio-Temporal analyses to identify customer trends. A LightFM-based recommendation engine was also created to improve customer experience and business growth.

Forest Fire Analysis

Analyzed the impact of meteorological factors on forest fire spread in Portugal, using techniques for variable interaction, outlier identification, transformation, and selection. This led to a statistical model offering insights into forest fire environmental dynamics.

CartFuel - AI Enabled Cart Conversion

CartFuel is an analytics driven system aimed at reducing eCommerce cart abandonment by predicting user behavior and increasing consumer confidence with strategic product information and reviews.

Amazon Reviews Sentiment Analysis

Performed sentiment analysis on Amazon.com reviews using text processing and machine learning algorithms, achieving 84% accuracy. This analysis offers insights into customer opinions, aiding in understanding consumer satisfaction and preferences.

Application Development

MediBase

Developed a secure, decentralized EMR system with a hybrid architecture and biometric authentication, ensuring medical record integrity. Showcased at GE SynerGE Hack'e'lth 2019 in Bangalore, it demonstrates potential in transforming healthcare data management.

Farmer's Buddy

A multi-platform mobile application for connecting Farmer's around India with other farmers and agriculture experts to provide informative and research based support for better agriculture needs tailored to their specific farms.