Bhanu Prasad Cherukuvada
I'm a software engineer at PayPal, where I work on Pal, an agent runtime and harness.
Here, I build small, independent experiments in sessions, model context, sandboxed tools, and recovery.
Featured experiment
What does the model actually receive?
Inspect the boundary between saved history and the next request.
Read the experiment
Latest writing
All postsWhy might cheaper AI lead to more AI work?
A working mental model connecting unit cost, useful tasks, measurement and changing bottlenecks.
Why do LLMs generate one token at a time?
How choosing each token creates a dependency, how speculative decoding checks several proposed tokens together, and when that trade saves time.
How does DeepSeek share its KV cache?
Follow active state, retained storage and the extra restore path that DualPath adds.
Did we actually reuse the previous turn’s cache?
Compare consecutive prefixes, usage denominators and the charges created when output returns as input.
Can we keep a cache warm—and is it worth it?
Separate retention policy, matching traffic, warm-up calls and total session economics.
Selected work
Pi (Bhanu Cloud)
2026A personal TypeScript agent runtime with PostgreSQL-backed sessions, Discord and CLI entry points, and a durable response outbox.
TypeScript, PostgreSQL, Agent runtimes
Agent runtime lab
2026Small, reproducible experiments in saved history, model context, persisted sessions, sandboxed tools, and retries. Runnable examples and the evidence behind the writing.
TypeScript, Agents API, Experiments
typing.app
2025Typing practice app with auth, subscriptions and stats. Built on Next.js, Better-Auth and Stripe.
Next.js, Better-Auth, Stripe, Drizzle