Hello, I am Chirag

I go by Srce Cde (srcecde) on the internet. I build things, then explain how they actually work - through blogs and YouTube videos (20k+ subscribers). My background spans ML, cloud, databases, and DevOps, with a current focus on LLMs, RAG, and agentic AI.

Here you'll find hands-on engineering guides, the math behind ML explained with intuition, and cloud walkthroughs you can follow along with - plus open-source code for most of it on GitHub. Start with the latest articles and videos below, or subscribe to the newsletter to get new posts in your inbox.

YouTube GitHub LinkedIn X Reddit

Temperature doesn't make your model creative

Temperature doesn't add creativity. It controls how far the model samples into variety it already had. Temperature, top-k, and top-p from first principles, with the proof that ranking never changes

llmtemperaturetop-p

Jul 22, 2026 · 21 min read

Drop one tool from your request: one GPT-5 version keeps 76% of it cached, another keeps nothing

Drop one tool from an identical request: gpt-5.2 keeps 76% of the prompt cache, gpt-5.5 keeps nothing. Fifteen replicated runs per version, with raw data.

prompt cachingllmopenai

Jul 8, 2026 · 16 min read

Reciprocal Rank Fusion (RRF): how it works and when to skip it

How Reciprocal Rank Fusion (RRF) merges retriever results in hybrid search, why it beats averaging scores, and the tradeoff that tells you when to skip it.

reciprocal rank fusionrrfhybrid search