Anand Mohan

Applied Scientist, Amazon AGI

Advancing Speech-to-Speech LLMs and continual learning for next-generation AI systems.

anandmoghan — zsh

Last login: on ttys000

anandmoghan@universe:~$ _

Last login: on ttys000

anandmoghan@universe:~$ echo "Hello World!"

Hello World!

anandmoghan@universe:~$ whoami

Applied Scientist at Amazon AGI

Specializing in Speech-to-Speech LLMs, post-training speech LLMs and incremental learning

anandmoghan@universe:~$ cat about.txt

I lead research on Speech-to-Speech Large Language Models at Amazon AGI.

My work spans post-training speech LLMs, production ASR systems, federated learning frameworks,

and multilingual speech recognition that powers AI at scale.

With 7+ years at Amazon and Masters in AI from Indian Institute of Science,

I've published 6+ papers in top conferences like ICASSP and INTERSPEECH.

anandmoghan@universe:~$ ls projects/

speech-to-speech-llms/

post-training-speech-llms/

incremental-learning/

multilingual-asr/

federated-learning-framework/

anandmoghan@universe:~$ _

Applied Scientist at Amazon AGI specializing in Speech-to-Speech LLMs, post-training speech LLMs and incremental learning.

With 7+ years at Amazon and M.Tech in AI from Indian Institute of Science, I've published 6+ papers in top conferences like ICASSP and INTERSPEECH.

Based in United Kingdom, working on the future of AI communication.

Flow Matching for Speech Generation

From diffusion to flow matching: how modern TTS models are trained, how they sample fast enough to stream, and where they still break.

From RLHF to reasoning models: how RL techniques align and enhance modern LLMs.

A deep dive into attention mechanisms powering modern LLMs.