Gunjan Dhanuka

Hello, I’m Gunjan.

I’m an AI Research Scientist at Mistral AI, working on multimodal models. My current work spans OCR and document understanding, as well as computer-use capabilities. I was a core member of the team behind Mistral OCR 4 and Mistral OCR 4.1, state-of-the-art OCR models for document intelligence. I care about making models genuinely helpful with everyday tasks: understanding documents, navigating interfaces, and turning human intent into useful actions.

I completed an MS in Computer Science at Carnegie Mellon University and a B.Tech. in Computer Science and Engineering at IIT Guwahati. At CMU, I worked on compression-based methods for LLMs and memorization in diffusion models with Zico Kolter and Pratyush Maini, and on model debiasing with Fernando de la Torre and Shayok Chakraborty. Before that, I interned as an ML Engineer at Apple and as a Software Engineer at Rubrik.

I enjoy turning fuzzy problems into concrete experiments, building the infrastructure to test them, and writing down what I learn along the way. My work on weakly supervised video anomaly detection was accepted to WACV 2025; recent work on compression-based memorization in diffusion models was accepted to the ICML MemFM Workshop 2025.

Outside of work, I’m usually taking photos, planning a trip, playing a game, or watching an improbably specific engineering video on YouTube. I’m also starting to collect trail notes and trip reports.

Beyond the screen

The things that keep my curiosity well-fed.

Some problems are best solved at a desk. The rest tend to get better after a long hike, a close football match, a beautifully designed game, or an unnecessarily deep coding rabbit hole.

mountain hours trail maps football nights game worlds camera roll

Follow the trail notes ↗