I am an AI Engineer with 5 years of experience, skilled in deep research, building end-to-end ML pipelines, and scalable AI solutions.

• Current Focus: Developing a multimodal No-Code Agentic AI platform at IBM Labs, utilizing LLMs, VITs, and Reinforcement Learning. This work currently reduces incident resolution time for IaaS teams from 24 hours to under 1 hour through intelligent triage and self-healing agents.
• Expertise: Agentic AI frameworks, Reinforcement Learning for model alignment (RLHF, PPO), Multimodal systems, and RAG pipelines.
• Experience: Software Developer II at Johnson Controls, where I co-developed a patented NLP+CV pipeline for automatic incident detection. I also received the "Pat on the Back" award 4 times for exceptional engineering contributions.

My research interests primarily revolve around developing autonomous, aligned, and efficient AI systems.

• Agentic AI & Reinforcement Learning (RL):

  • Leading research on autonomous agent behavior targeting critical automation use cases across various verticals.
  • Extensive work in developing Agentic flows and using Reinforcement Learning for model alignment.
  • Researched LLM-based agents optimized using RLHF, PPO, and epsilon-greedy, focusing on capabilities for self-reflection and deep tool reasoning to build scalable autonomous AI systems.
  • Developed and deployed a modular Multimodal Agentic AI framework that improved flexibility and reduced response latency by over 80% in incident resolution workflows.
  • Relevant Coursework: Deep Reinforcement Learning.

• Generative Models & Recommender Systems (RecSys):

  • Currently developing a domain-specific foundation model for E-Commerce personalization using a BERT-based architecture and Contrastive Learning.
  • Exploring novel pretext tasks and custom loss functions to improve semantic representation and pretraining efficiency.
  • Fine-tuned DistillBERT on the AmazonReviews23 dataset for multi-label classification, achieving a hit rate of 0.7 using TF-IDF and confidence thresholding.
  • Built a movie recommendation app (Moviemate) using collaborative filtering algorithms.
  • Research Interests: Recommender Systems, Agentic Patterns, and Generative Models.
  • Relevant Coursework: Recommendation System, Self-Supervised Learning.

• Alignment & RAG:

  • Contributed to the open-source InstructLab framework, fine-tuning LLMs for education by aligning them to domain-specific datasets.
  • Boosted chatbot response accuracy to 0.8 and relevance score to 0.9 by developing a Retrieval-Augmented Generation (RAG) pipeline.

• Reinforcement Unlearning:

  • Exploring different existing unlearning algorithms like sample space shrinking, poisining etc to understand how unlearning is done also exploring philosophically how unlearning can be achieved.
  • Exploring novel ways to do unlearning using Measure Theory, Stochastic Calculus, Martingales etc.

• Master of Technology in Artificial Intelligence
  • International Institute of Information Technology, Bangalore (IIIT-B)
  • Jul 2023 - Jul 2025
• Bachelor of Technology in Computer Science
  • Netaji Subhash Engineering College, Kolkata
  • Aug 2014 - Jun 2018

I am also a contributor and a co-author in the patent titled "Building system with automatic incident identification": [Link to Patent]

• I am ambidextrous.
• I know how to play 4 different musical instruments (Guitar, Ukulele, Harmonica, Mandolin).
• I received the "Pat on the Back" award 4 times at Johnson Controls for exceptional engineering contributions.

• Email: rsunnidhya@gmail.com