AI-Powered Automations and LLM Integrations in Energy SectorDeveloping a RAG chatbot for improved Smart energy meter troubleshooting.

Researched energy efficient AI model training and inference techniques with Prof. Carlo Lipizzi.

Teaching Assistant for "Mathematics for Machine Learning" Course:Facilitating student understanding in mathematical concepts crucial for machine learning, including linear algebra, probability, and statistical theory, through engaging tutorials and one-on-one mentoring sessions.

Assessed and supported graduate students in python coursework through meticulous grading and constructive feedback.

• Enhanced the BLEU score for Adverse Effect Predictions on medical data by 15% to 0.85 through fine-tuning of Language Models, including T5 and GPTJ, helping a company extract side-effects from medicine reviews.• Developed a Mixture-of-Experts Model based on three LLMs, achieving over 95% accuracy in extracting 10+ personal Information points at once from real-time customer care conversations.• Built a custom Python package and a web interface to autonomously setup environment… Show more • Enhanced the BLEU score for Adverse Effect Predictions on medical data by 15% to 0.85 through fine-tuning of Language Models, including T5 and GPTJ, helping a company extract side-effects from medicine reviews.• Developed a Mixture-of-Experts Model based on three LLMs, achieving over 95% accuracy in extracting 10+ personal Information points at once from real-time customer care conversations.• Built a custom Python package and a web interface to autonomously setup environment and train ML models on a supercomputer(NVIDIA A100s). Uses Github Webhooks to pull new code securely and add it to a execution queue which zeroed machine’s ideal time saving money and doubled the throughput increasing team’s efficiency. Integrated with RSA encryption, Redis database, WandB monitoring, and AWS S3 storage for seamless operations.• AI Interviewer: Led the development of an AI Interviewer by pretraining and fine-tuning a Speech-to-Text model on 55,000 hours of Indian English data. Implemented speaker identification, QA pair extraction and embeddings with LLMs. Designed a live interviewer with GPT-4 using intelligent prompt engineering capable of dynamic question generation and final candidate assessment based on their response and job description.• LLM and Prompt Engineering Training Sessions: Conducted training on GPT-3/4 for the entire AI/ML team at Cognizant, covering use cases, fine-tuning methods, and business implementation strategies. Followed by sessions on advanced Prompt Engineering techniques such as AutoGPT, HuggingGPT, Voyager, Mindstorm, and 20 other methods, supported by my own articles, research papers, GitHub code walkthrough and analysis. Show less

• Deployed a pipeline to clean and process data using Spark, visualize it using Tableau and store it in Azure cloud.• Developed an NLP model to classify news articles into 45 genres achieving an accuracy of 92%.

• Created an Alexa skill for teaching youngsters various educational concepts through voice intercommunication.• Created a whole plot in which users may engage and learn about various nations' capitals.