RESEARCH & EXPLORATION
Our research connects India-focused multimodal language models, vision-language adaptation, quantization and efficient serving with vLLM.
Vision-language adaptation
BharatGoAI focuses on India-focused multimodal models that connect vision and language. Our work includes LoRA fine-tuning and GRPO/GSPO-based adaptation for Indian-context tasks, including handwritten and printed document understanding. Indian-language text and task-specific evaluation are part of the same research direction, alongside practical deployment requirements.
Model quantization
Our model quantization work includes INT8, FP8, W4A16 and NVFP4, with deployment through vLLM. We focus on the relationship between model representation, memory use and practical serving. The project collection links to quantized artifacts and identifies the stage of each model or deployment effort.
Efficient serving with vLLM
BharatGoAI connects model adaptation and quantization with efficient serving through vLLM. The deployment direction includes accessible hardware and accuracy evaluation across hardware. Model artifacts and deployment work are listed separately so readers can distinguish released quantization projects, research experiments and work that remains in development.
OUR ENGINEERING PERSPECTIVE
A useful framework for thinking about
AI/ML projects, from start to finish.
Understand the intended user, the task and the constraints.
Investigate models, data and approaches that fit the problem.
Compare outcomes and examine quality and resource trade-offs.
Use what the experiment reveals to shape the next iteration.
FOLLOW THE WORK
Find the founder’s model work on Hugging Face. Refer to individual project pages for their documentation, status and licence.
Research direction by Lokesh E, Founder & AI/ML Engineer. Explore the multimodal model and quantization projects or discuss research collaboration.
THE NEXT CONVERSATION
Have a research idea or an AI/ML project in mind?
We’d love to hear what you’re thinking.
Last updated: