Shlok Khemani is a writer and programmer investigating personal AI, memory and agent interfaces. His public projects turn analyses of assistants into local software for email, reminders and ongoing work.[1]
Start with Lessons from Studying Every Memory System →
GitHub profile 
X · shloked Career
Khemani’s owned biography describes earlier work at Decentralised.co, writing stories and building cryptocurrency products. His current interests center on assistants that retain context, interfaces between people and models, and agents that keep working outside a single conversation. His writing and public implementations let readers compare a product’s observed behavior with a working design.[1]
Contributions
OpenPoke implements a conversational assistant with a FastAPI backend and a Next.js interface. Separate interaction and execution components handle the exchange and the work; a scheduler and background watchers support reminders and email alerts. Gmail actions use Composio and model calls use OpenRouter. The repository makes the required provider keys and account setup explicit, allowing developers to inspect the orchestration instead of treating the assistant as an opaque chat window.[3]
Ideas
His memory-system comparison distinguishes saved facts, background synthesis into a user profile and retrieval from previous conversations. A profile can supply context without a person repeating it, but it can preserve stale assumptions; retrieval can recover an earlier exchange while increasing serving work. He argues that useful memory also needs to learn what changed outside the chat. The architectures are observed product designs, with choices shaped by the assistant’s intended interaction.[2]
Tools, writing & talks
Writing, research & talks
Start here
Lessons from Studying Every Memory Systemai.engineerA comparison of saved facts, synthesized profiles and retrieval, including their maintenance and cost tradeoffs.
Begin with the saved-fact, synthesized-profile and retrieval examples. No coding is needed to follow the comparison; when trying a system, inspect what it remembers and how you can correct an outdated assumption.
Presenter[2]A year of reverse-engineering consumer memory systems, opened by a ChatGPT profile recording a trip to Turkey the author never took, and the observation that nothing notices the conflict though the evidence sits in his email.
Catalog workOriginal source