Harsh Gupta Email & Phone Number

London, United Kingdom

Columbia, South Carolina, United States

Working on adversarial attacks on Large Language Models (LLMs) and their defenses.

Bengaluru, Karnataka, India

Working on building agentic systems

Bengaluru, Karnataka, India

Developed personalized AI-driven career solutions for teenagers using LLMs. Migrated from Langchain framework to custom framework using Pydantic and FastAPI and reduced the latency by 20x by enabling streaming. I also worked on implementing RAG over knowledge graph using LlamaIndex and NebulaGraph and finetuning Mistral 7B using PEFT and LoRA.

Hyderabad, Telangana, India

Awarded Pre-Placement Offer (PPO)Created the Auri-Search Portal from scratch, which is an AI-enhanced search engine with predictive capabilities for bio-physical properties using Transformer-based models

Birmingham, England, United Kingdom

Worked on Reinforcement Learning Strategies in 2-Player Iterated Prisoner Dilemma Games. Currently, the paper is under review "Cooperation through Combining Evolutionary Dynamics and ReinforcementLearning in Networked Feedback-Evolving PD Games", Under the guidance of Prof. Leonardo Stella.

Ahmedabad, Gujarat, India

• Understanding 10-K narratives across various life cycle stages of firms, Under the guidance of Prof. Neerav Nagar
• Proficiently extracted and parsed the 10-K Annual Reports for 18,000+ U.S Companies from 2010 to 2020 into Item1, Item 1A, and Item 7
• Computed the McDonald Loughran Sentiment, i.e. a Lexicon-based sentiment analysis with 80,000+ words and implemented POS tagging
• Performed sentiment analysis of long reports by stacking chunks in FinBERT and compared the performance with McDonald Sentiment