Amrit Gupta
Development
MA, United States
Skills
Machine Learning (ML)
About
Amrit Gupta's skills align with IT R&D Professionals (Information and Communication Technology). Amrit also has skills associated with Programmers (Information and Communication Technology). Amrit Gupta appears to be a low-to-mid level candidate, with 3 years of experience.
View more
Work Experience
Artificial Intelligence Intern
Prompt Tuning, LoRA, RLHF, Streamlit
July 2023 - December 2023
- Drove 5x improvement in answer completeness evaluation to enable guardrails for helpdesk AI by fine-tuning state-of-the-art LLaMA 2 language model on synthetic data, achieving industry-leading GPT-4 performance levels. • Engineered a personalized newsletter generation pipeline using retrieval augmented generation (RAG) to deliver tailored content aligned with individual client investment profiles and portfolio insights. • Represented Fidelity in NeurIPS Large Language Model (LLM) Efficiency Challenge 2023. Fine-tuned LLaMA 2, Mistral models in a Mixture of Experts (MoE) architecture using PEFT & QLoRA, achieved 10% performance increase in HELM benchmark tasks. • Enhanced the functionality of Morphi, an internal package for the organization-wide deployment of state-of-the-art LLM's fine-tuned on domain-specific data to optimize performance for targeted natural language processing tasks.
Machine Learning Engineer
LLMs, BERT, DDP, DeepSpeed, FastAPI
September 2020 - August 2022
- India • Created a model for qualitative ranking of candidate summaries. Automated the summary evaluation process, reduced evaluation time by 40%. • Improved quality of domain-specific summaries by augmenting input with external information derived from knowledge graphs. Achieved 23% performance increase measured by BERTScore. • Developed aspect-specific evaluation models for summarization using synthetic data for automating the model validation process, reduced model evaluation time by 50%. • Spearheaded the development & integration of the above projects in TCS Trustworthy AI framework.
Machine Learning Intern
Honeywell
January 2020 - July 2020
- Developed POC using Machine Learning & Deep Learning techniques for Octane Number Prediction using Near- Infrared Spectroscopy data, reduced time for prediction by 30% • Deployed POC using ReactJS & Python Flask. POC used for customer demo and for building full application. Followed Agile Development Methodology.
Summer Intern
Python, Bash, Linux, RHEL
May 2019 - June 2019
- India • Automated Workflow of Unmanned Aerial Vehicle - Real Time Simulator reducing time taken for start up and operations by 50%.
Education
Northeastern University
Master of Science in Computer Science
September 2022 - May 2024
