Featured Research
Tokenizers are All we Need
A grounded walkthrough of how modern tokenizers work, why they matter, and how their hidden mechanics shape efficiency, language handling, and real-world model performance.
Why India Needs Its Own LLMs
The future of AI in India depends on building language models that understand our culture, context, and languages.
Diving Deep into the Mamba Model Implementation
A technical exploration of implementing the Mamba model in PyTorch, breaking down the architecture, key components, and challenges.
All Research
The Smartest Person in the Room
Coffee coffee coffee
Before the Code
I saw the diff b/w two brain types,
Observation Writes First
The one who remembered more syntax, more commands, more structures, was seen as the better developer.
LLMs Counting Digits, Not Storing a Number Line
LLMs Counting Digits, Not Storing a Number Line
When LLMs Multiply!!
When LLMs Multiply: Right Ballpark, Wrong Answer (and What the Hidden States Actually Say)
Contribute to Bhasha's Research
Join us in exploring the future of language and technology.
Contact Research Team