Rohan Rao

I started in hardware — an Electrical Engineering dual degree from IIT Madras, where I also co-founded Dynamove, an ADAS startup building intelligent dashcams for Indian trucking fleets. That went through Axilor Ventures and reached the Top 50 at IIGP 2018, which was enough to convince me that building real products on top of cutting-edge research was worth doing full-time.

From there I went to CMU's Robotics Institute for a master's in Robotic Systems Development, then joined NVIDIA where I spent several years on autonomous vehicle software — working on the NVIDIA Xavier SoC and the DRIVE AV platform with Mercedes-Benz. In 2024 I made the shift to generative AI, which is where most of the interesting unsolved problems now live.

My current work at NVIDIA sits at the boundary between research and production: translating LLM capabilities into things that actually run reliably at scale. That means RAG systems, knowledge graph pipelines, NIM microservices, and agentic architectures. I published HybridRAG — a paper on combining vector retrieval with graph-structured context — with colleagues from NVIDIA and Citadel at ACM ICAIF 2024. I also maintain an active technical blog and have spoken at Graph the Planet 2025 on GPU-accelerated GraphRAG.

Timeline of You started as a hackathon project at the LlamaIndex Agentic RAG-a-thon 2 in October 2024 — we won 1st place Grand Prize out of 454 participants. The original prototype used agentic RAG to ingest career evidence from multiple sources, build a knowledge graph, and generate EB-1A petition documents. It's been a product since. Separately, I won 1st place at the Stanford LLM x Law Hackathon in April 2025 with JurisLink, a tool that maps conflicts of interest across legal databases and corporate structures for international arbitration. Same underlying instinct: AI is most useful when it's doing structured reasoning over messy real-world data, not just generating text.