Skip to content

Blog

Notes & essays

Occasional notes on research, programming and the road here.

AI14 min read

Your LLM Has Two Memories. You’re Only Using One.

Classic machine learning kept learning: when customers changed, we retrained. Most LLM systems freeze the weights and push every lesson into the prompt. RAG, MCP, agents, memory: take away the names and they all do the same thing.

AI7 min read

LLMs Are Conditional Probability Machines

“Next-token predictor” is technically true, but it only tells half the story. The word that matters is conditional — and it describes what you control.

Coding2 min read

yield in Python

Understanding the yield keyword in Python — how generators work, when to use them, and their key advantages over regular functions.

Coding2 min read

Circular Queue: Python Implementation

A clean Python implementation of the circular queue (ring buffer) data structure that beats 94% of LeetCode submissions.

Experience2 min read

My Experience at Heidelberg Laureate Forum 2019

How a last-minute application led to one of the most memorable experiences of my academic journey — meeting Yoshua Bengio and Jeff Dean at the Heidelberg Laureate Forum.

Mathematics2 min read

Mutually Exclusive vs. Collectively Exhaustive

A quick primer on two foundational concepts in probability theory that are often confused — mutual exclusivity and collective exhaustiveness.