Excited to share that 2 papers have been accepted to ICLR 2026 🇧🇷! These span over the two core themes of my PhD: safe and controllable adaptation of ML models.
I’m also graduating this semester, happy to connect about related opportunities!
Longer posts soon, but TL; DRs👇
Final-year PhD candidate @GeorgiaTech. Previously undergrad @IITDelhi. Controllable Generation. Adversarial Robustness. Graphs. Language. Diffusion.
- How to flexibly adapt RAG systems to any domain without retraining or arbitrary document restructuring? We turn to the OG method of domain-specific ontologies and present Ontology-Grounded (OG)-RAG at #EMNLP2025 (joining virtually 11/04 7-8 pm ET). (1/N)
- Had a great time presenting the poster here at #ICML2024 yesterday! TLDR: We present PRODIGY to generate graphs from any diffusion model while satisfying arbitrary constraints..@GeorgiaTech researchers have unlocked diffusion models for custom drug design, materials discovery, and social network analysis! Check out PRODIGY, presented this week at #ICML2024! @gtcomputing @GTResearchNews @srijankedia @ksartik @icmlconf cc.gatech.edu/news/new-machi…
- To my tennis family and beyond, With Love, Roger
- One piece fandom definitely has better data visualizers than half of us data scientists.




