I build LLM-driven recommendation systems for the planning domain—turning vague goals and hard constraints into actionable plans.
Methodologically, I work across ML / NLP / DL: fine-tuning, RAG/agents, and preference modeling.
My focus is constraint-aware planning (time, budget, policy) and multi-objective optimization over quality, cost, and latency.
I develop evaluation stacks—rubric/judge models, offline metrics, and human-in-the-loop studies—for reliable comparisons.
Reliability matters: guardrails, uncertainty estimation, and hallucination mitigation to keep recommendations safe and stable.
Goal: bridge methodology → deployment, shipping planning systems that are measurable, auditable, and genuinely useful
Abstract: TripTide focused on how different LLM models performs based on real-life disruptions in Travel Planning Scenario. We develop 3 novel evaluation metrics to score how the models perform.
Submitted : WWW Conference (A*), 2026
Abstract: TripTide focused on how different LLM models performs based on real-life disruptions in Travel Planning Scenario. We develop 3 novel evaluation metrics to score how the models perform.
Submitted : WWW Conference (A*), 2026