Building Frontier Multimodal Multilingual AI with Efficient Research
Building frontier multimodal and multilingual AI systems is no longer just about scaling models. It requires careful data design, efficient training pipelines, rigorous evaluation, and fast research iteration under real world constraints.
In this talk, Adithya will present his work on the M3DR paper as a case study for building a multilingual multimodal document retrieval system.
Using M3DR as the anchor, he will walk through the full research lifecycle using Claude Code workflow, from problem formulation and dataset creation to model training, evaluation, and deployment oriented research decisions.
Beyond the paper, Adithya will cover how they optimised research workflows to move faster with limited compute. This includes practical strategies for multilingual data curation, multimodal evaluation, infrastructure design, experiment tracking, and open source driven iteration.
He will also share lessons from building one of India’s first Indic large language models and maintaining an Indic LLM leaderboard, highlighting what it takes to build frontier systems for underrepresented languages at scale.
Speaker Bio
Adithya S K is an AI researcher and builder working on multilingual and multimodal AI systems.
He is a Research Fellow at Microsoft Research and the founder of CognitiveLab, an open-source research lab focused on accessible AI.
He has built one of India’s first Indic large language models, created an Indic LLM leaderboard, and led multiple open source projects with over 10k GitHub stars.
Speaker Socials
GitHub https://github.com/adithya-s-k
LinkedIn https://www.linkedin.com/in/adithya-s-k
To attend online:
Add to calendar: https://shorturl.at/fgFQx
Gmeet link: meet.google.com/vtk-fjkp-pbu
Pre read material
M3DR paper https://arxiv.org/abs/2512.03514
M3DR HTML version https://arxiv.org/html/2512.03514v1
M3DR on Hugging Face https://huggingface.co/papers/2512.03514
Indic LLM Leaderboard blog https://www.cognitivelab.in/blog/introducing-indic-llm-leaderboard
Indic LLM Leaderboard news coverage https://analyticsindiamag.com/ai-news-updates/cognitivelab-releases-indic-llm-leaderboard
The goal of this session is to share actionable insights for researchers and builders who want to translate cutting edge ideas into scalable and usable AI systems.