Krishnakanth Alagiri

A nodel graph-based Retrieval-Augmented Generation system to improve LLM performance in social reasoning tasks, achieving significant accuracy improvements by integrating domain-specific knowledge and advanced prompting techniques.

Implemented an Alexa Bot leveraging Articulatory Phonetics, weighted Levenshtein, and BERT for pun recommendations. Successfully developed sound and sense similarity using a robust database, enhancing user engagement and satisfaction.

AI internet moderation system using BERT and RESNET50, achieving state of the art performance in toxic text classification and image classification. Built scrapers for preparing a sensitive image classification dataset.

An unified system for certificate generation and distribution, volume mailing, advanced form generation, mailing lists, monitor databases and a link shortener for small organisations/clubs.

Leveraged BERT for multi-class sentiment analysis to dynamically track and predict emotional shifts across text message sequences, utilizing a vector database for enhanced data handling and accuracy.

Built a translator using Convolutional Neural Networks (CNN) to classify signs with over 86% accuracy. The application was able to understand the variations among different national and regional sign languages dialects through crowd-sourcing.