About
I work on agentic AI at LinkedIn. Day to day that means evaluation, context engineering, and the unglamorous work of getting agents to hold up once real people start using them. Before this I worked on LLM inference and recommender infrastructure.
I studied computer science, mathematics and statistics at Amherst College, and wrote two honors theses — one formalizing chip-firing and Riemann–Roch for graphs in Lean 4, the other a Python package for exploratory analysis of multivariate discrete data.
What this site is for
Writing, mostly. Things that are too long, too technical, or too permanent for a feed post. If you want the current version of my job, my full background, or my day-to-day thinking, LinkedIn is the better place — it’s where I actually post.
- Papers and citations → Google Scholar · ORCID
- Code → GitHub
- Short takes → X
- Everything else → ddmavani2003@gmail.com
Open source
- ccrvam — model-free exploratory analysis of multivariate discrete data with an ordinal response
- chipfiring — chip-firing games on multigraphs
- QuantileFlow — unified quantile sketching for streaming anomaly detection
Elsewhere
A few pieces I’ve written or been quoted in:
- Lean4: how the theorem prover works and why it’s the new competitive edge in AI — VentureBeat
- The teacher is the new engineer: inside the rise of AI enablement and PromptOps — VentureBeat
- Accelerating LLM inference with speculative decoding — LinkedIn Engineering Blog
- I landed a job at LinkedIn by posting on the platform — Business Insider
- The real AI bubble is in data centers no one can power up — HackerNoon
- From correlation to checkerboards: model-free EDA for categorical data — HackerNoon
Off-hours
Badminton, archery, and talking people into bungee jumps.