🔬 Research & Publications

Graduate Student Researcher — BioNLP Lab @ UMass Amherst

Guide: Prof. Hong Yu | August 2025 – Present

• Designed an ontology-driven self-play framework where dual LLM agents (Q&A) co-evolve to fill biomedical knowledge gaps leveraging ontology data, achieving 12% performance boost over Supervised Fine Tuning.
• Implemented post-training alignment techniques like DPO, evidence-aligned RL and enhanced factuality by 8%.

Publications & Research Projects

AI-Generated Lecture Slides for Improving Slide Element Detection and Retrieval
ICDAR 2025 | Aug 2024 – Feb 2025
[Paper] [Project Website]

Led research on an LLM-driven synthetic slide generation pipeline. Trained models for slide layout detection and image retrieval, delivering SOTA performance.

Python PyTorch LLMs Computer Vision

Projects & Hackathons 🏆

💼 Professional Experience

Advanced Data Science Associate — ZS Associates

Pune, India | November 2024 – July 2025

• Developed and deployed a Switch Prediction Model to predict patients switching from ongoing drug treatments using real-world data and ML/DL models (Bayesian classifier, XGBoost, ANN, TabNet).
• Delivered 70% precision @ 60% recall; created visualizations, communicated results and SHAP analysis to clients, aiding in optimizing marketing efforts with projected cost savings of up to $500K.
• Built patient cohorts using SQL and performed patient archetyping via clustering.
• Engineered a dashboard to visualize model results and automate retraining and inference, enabling one-click MLOps.

AI/ML Engineer — Decimal Point Analytics

Mumbai, India | July 2023 – October 2024

• Designed a predictive maintenance pipeline to forecast defects in industrial machinery using time-series models.
• Mentored two interns on a Knowledge Distillation and Few-shot Object Detection project.
• Built a RAG system for Q&A over PDFs/Excel files using Pinecone, integrating knowledge graphs to get a 9% accuracy boost; used multi-threading to reduce latency by 75%.
• Created an agent-based hybrid web scraper by finetuning Llama2 (7B/13B) via LoRA for text extraction and using GPT-4 for code generation, reducing downtime from 12 hours to 3 minutes.
• Developed a DALLE-3 pipeline to generate fabric design patterns from textual specifications.
• Prototyped multiple GenAI products using open/closed LLMs, FastAPI, Streamlit, and Django.

AI Research Intern — Co-op Program — Raapid AI x IIT Jodhpur

Remote, India | Sept 2022 – May 2023

• Engineered a table detection pipeline by finetuning CascadeTabNet and DIT, achieving 76% mAP.
• Applied novel augmentation strategies, improving mAP by 7% on a private gold-standard dataset.

AI/ML Intern — Decimal Point Analytics

Remote, India | May 2022 – July 2022

• Built an end-to-end accident detection and damage segmentation pipeline for CCTV footage.
• Conducted experiments with Vision Transformer, YOLO, and other architectures, securing an F1 score of 87%.

🛠️ Technical Skills