At TCS Research, I revived an inherited LLM-based knowledge-extraction framework for web and legal documents, restoring its executability and diagnosing end-to-end failure modes. I designed a fault-tolerant multi-agent orchestration layer with checkpointing and automatic retry and resume logic.
I built ontology-guided pipelines for extracting entities, relations, and events, using Owlready2 and Pydantic for structured validation. I also developed iterative extraction and question-answering pipelines with MiniLM retrieval, evaluating performance with F1 and QA accuracy.
I co-authored and submitted a paper to ACM IKDD CODS 2026 on task-driven schema evolution for ontology-guided knowledge graph extraction, contributing to problem formulation, baseline evaluation, and manuscript development. My projects also include a RAG-based legacy system documentation assistant and deep-learning models for solar power forecasting.
