Building useful AI for the real world.
With a bias toward clarity.
I work on recommender systems, sequence modeling, foundation models, and reliable evaluation. This is my corner of the internet for things I build, learn, and share.
A practical research mindset.
I like the space between a paper and a product: understanding how a method works, implementing it carefully, then finding where it creates real value.
Personalization
Recommender systems and sequence models for better experiences at scale.
Foundation models
Tools and abstractions that make profile and event data useful for modern ML.
Reliable evaluation
Measuring what actually improves an AI system—not just what looks impressive.
Things I’ve shipped.
Open-source experiments and tools built to make research easier to use.
fastpragma
A Python API for PRAGMA-style foundation models—from tokenization and pretraining to embeddings and fine-tuning.
llm-rsa
Recursive Self-Aggregation for LLMs: sample and synthesize candidate answers into a stronger response.
“The best work sits where curiosity meets usefulness. I’m still learning, building, and writing my way there.” — Risheekkumar