About Belgavi.AI Lab

A personal technical reference built by Sandeep Belgavi Ashok Kumar, a Senior Engineering Manager and Architect working at the intersection of scalable distributed systems and modern AI.

What this site is

Belgavi.AI Lab collects deep technical material in two complementary forms: written explanations that aim for concrete numbers and working examples over marketing claims, and interactive single-file labs that run in a browser with no install, no build step, and no JavaScript framework dependency.

Topics span LLM internals, agent protocols (MCP, A2A, ADK, AP2), Apache Cassandra, distributed consensus, system design, networking, security, observability, and the realtime media stack (WebRTC, HLS, DASH).

Who writes it

The site is written and maintained by one person. Article drafts across the reference sections are produced with the assistance of large language models and then reviewed and corrected before publication; the interactive labs, the transformer math derivations, and the editorial direction are hand-built. Where a page is thin or still being expanded, it is marked as such rather than presented as a finished treatment.

If you find an error, a correction is genuinely welcome and will be fixed — see the contact page.

Why it exists

Most explanations of these systems stop at the level of a diagram. The intent here is to go one level lower: show the arithmetic, show the failure mode, show the mechanism running. The labs exist because some ideas — attention head behaviour, consensus under partition, quantization error — are far easier to understand when you can move a slider and watch the numbers change.

Contact

Questions, corrections, collaboration, or speaking: get in touch. Typical response time is 2-3 business days.