Welcome to Yaswanth’s Homepage!

I am a graduate researcher who is working on Language Models and I did my undergrad and masters from the Center for Excellence in Artificial Intelligence , Indian Institute of Technology Kharagpur. I was also an undergraduate researcher at the Complex Networks Research Group, IIT Kharagpur under the supervision of Prof. Pawan Goyal for the past 3 years. If you have any ideas that you would like to collaborate on, hit me up!

Publications 📚

• Review-Feedback-Reason (ReFeR): Improving Evaluation and Reasoning through Hierarchy of Models
  First Author Long Paper, Under Review at ICLR 2025 (Accepted at 2 NeurIPS 2024 workshops)

• DepNeCTI: Dependency-based Nested Compound Type Identification for Sanskrit
  First Author Long Paper, Accepted at EMNLP 2023 (Findings)

• II-Bench: An Image Implication Understanding Benchmark for Multimodal LLMs
  Accepted as Poster in NeurIPS Workshop 2024

• VideoScore: Building Automatic Metrics to Simulate Fine-grained Human Feedback for Video Generation
  Accepted at EMNLP 2024

• DPO Kernels: A Semantically-Aware, Kernel-Enhanced, and Divergence-Rich Paradigm

Research Internships 🧪

News 🏆📚

• [Apr 2024] Waterloo Research: Started research internship at University of Waterloo’s TIGER Lab. 🎓
• [Dec 2023] Grant Increase: Microsoft Research doubled our grant to $40,000! 🎉
• [Oct 2023] Azure A100 Deployment: Launched A100 (80GB GPU) Server on Azure with 220GB RAM. 🚀
• [Oct 2023] EMNLP 2023: First author paper accepted at EMNLP 2023 (Findings). 🎉
• [Sep 2023] EMNLP 2023: Served as a Student Paper Reviewer. 🔍
• [Apr 2023] Best Thesis Awards: Received highest grades for both Bachelor’s and Master’s theses. 🥇
• [Apr 2023] Best Term Papers: Awarded for “Natural Language Processing”, “Graph Machine Learning”, “AI Design Lab”. 📝

Research Experience

🤖 FinAgent: Financial Trading Agent [Apr 2024 - Present]

• University of Waterloo: Developing a multimodal agent for enhanced financial trading and stock portfolio optimization. Created comprehensive dataset of financial news and stock prices for 100 companies.

📊 Automatic Evaluation Framework using LLMs [Aug 2023 - Apr 2024]

• Microsoft Research India: Developed a 3-stage evaluation framework with LLM agents (GPT-3.5, Mixtral-8*7b, Gemini-pro) for NLG evaluation. Achieved SOTA correlation with human evaluation.

📑 Event Extraction using Large Language Models [October 2023 - Present]

• University of Texas, Dallas: Leading a research project focusing on document-level and sentence-level event extraction with GPT-3.5, GPT-4 and various open source Large Language Models.

📚 Citation Integrity [June 2023 - September 2023]

• University of Illinois at Urbana-Champaign: Developed a new dataset, examined the integrity of citations, and classified them into categories, such as “irrelevant citation,” while also extracting evidence from the cited papers.

🌐 Machine Translation using Large Language Models [May 2023 - Sept 2023]

• University of California, Berkeley: Enhanced 500K lines of machine translation data with synthetic generation from LLMs, employing Google-OCR, BERT-based aligners, and Sandhi splitting techniques. Notably increased BLEU scores by 10% using the “No Language Left Behind” (NLLB) model and advanced post-processing methods.

🚦 Compressing Yolo Object Detection using NN-LUT [August 2023 - September 2023]

• Indian Institute of Technology, Kharagpur: Developed a pedestrian detection system with YOLOv4 architecture, trained on the EuroCity Persons dataset and compressed the model using Neural Network based Look up table for self driving cars. Also deployed a GRU model with autoencoder for intrusion detection using CAN dataset.

🧩 Nested Compound Parsing and Type Identification [December 2022 - April 2023]

• Indian Institute of Technology, Kharagpur: Introduced a novel task, dataset and framework focused on identifying correct parsing and semantic relations between components of compounds in Sanskrit.

💡 Commonsense Injection to Multimodal Reasoning Models [February 2023 - April 2023]

• Indian Institute of Technology, Kharagpur: Proposed a method to mitigate common sense mistakes made by the Multimodal Chain-of-Thought reasoning in language models by incorporating commonsense knowledge via knowledge graphs.

❓ Asking Clarifying Questions for Dialogue Systems [February 2023 - April 2023]

• Indian Institute of Technology, Kharagpur: Focused on Asking Clarifying Questions in an open-domain language system, encompassing two subtasks: determining when to ask a clarifying question and which question to ask.

🏷️ Semantic Tag Recommendation Framework for Quotes [August 2022 - November 2022]

• Indian Institute of Technology, Kharagpur: Predicted various relevant tags from a list of 39,000 available categories for a given Quote and used the context of each Quote as well.

🧠 Research Interests

• 🤖 Large Language Models
  • 🖼️ Multimodality
  • 🧐 Reasoning
  • 💭 Hallucinations
  • 🎯 Reasoning & Planning
  • 🎮 Agentic Behavior
  • ✅ Evaluation

✍️ Blog Posts

• Creating Automatic Knowledge Graphs from Pre-trained Language Models

🎉 Hobbies & Interests

• ♟️ Chess Enthusiast: I’m an avid chess player and have won multiple district-level tournaments. Fancy a game? Find me on chess.com and let’s play!
• 🏀 Basketball & 🏐 Volleyball: Whether it’s shooting hoops or spiking the ball, I love staying active and playing sports in my free time.
• 👨‍🍳 Cooking & 🌎 Travelling: Exploring new cuisines and cultures is my passion. I find joy in cooking up a storm in the kitchen and embarking on culinary adventures.
• 🧘‍♂️ Meditation: In the midst of life’s hustle and bustle, I find peace and rejuvenation through meditation. It’s my way of staying grounded and mindful.
• 🧢 High School Captain: I took on the mantle of leadership in high school, guiding and inspiring my peers as the school captain.