Reason fromfirst principles
AI should force you not to make mistakes.
I write, teach, and ship the standard I want tools to hold. The public artifact is arunveersingh/ai — 11 agent skills that refuse to proceed on a guess. I teach the same standard in English and Hindi. Tutorials stay in the library. Judgment stays here.
The standard and the talks
By the numbers
17years shipping production systems
11agent skills that will halt a bad plan
13selected talks on judgment, not syntax
Selected systems work
- AI skills that will not let you be wrong in peaceAI made producing work nearly free. Checking it, understanding it, and owning it did not get cheaper — so that is where things now go wrong.
- code-intel-mcpMCP server for AI agents — git lifecycle management, Zoekt code search, and cross-repo dependency analysis
- sbom-parserPython command-line tool that visualizes dependency trees from Software Bill of Materials (SBOM) files
- json-vs-protobufCompares the performance difference between JSON and Protobuf serialization.
Writing
Latency is a measurement. Slowness is a feeling. Mixing them up wastes the conversation.
Eliminating null in JavaNull is design-time uncertainty encoded at runtime. Patterns that take it out of the system.
The trouble with void methodsVoid methods hide outcomes. That makes tests brittle and intent harder to read.
Managing Null in the Spring Framework / Spring Boot: Advanced Strategies for Null-Free ArchitecturesColophon
39 public repos sit behind this page. It leads with the 4 that prove the standard — the rest is a lab, and lives on GitHub, not here.
Teaching happens on Arun | Engineering Judgment in English and Developer Notes Hindi in Hindi. The 13 talks worth arguing with are collected here. Written courses are listed here.
