Durgansh

I am an ai engineer,

I have been working in ai for around 4-5 years in various startups.

Majority of my works revolves around multi agent orchestration and conversational ai.

I care about the unglamorous parts that make agents work in production: evals, guardrails, cost control, access rules on every tool call, and clear traces of what each agent did and why. Some of my best lessons came from incidents, like an agent inventing teammates or a bot revealing its own tools, and the fixes that followed.

One example is a hotel concierge that answers guests across web chat, Instagram, Messenger and email, pulls answers from each hotel's own knowledge base, and books rooms through live property-management systems.

Another is a sales-performance platform where agents draft rules and write summaries, but every number is computed by the system, never by the model.

I care about the unglamorous parts that make agents work in production: evals, guardrails, cost control and clear traces of what each agent did.

I also enjoy low-level problems, like adapting the Claude Code SDK to non-POSIX filesystems and building a Flutter plugin for Apple hardware.

Based in London.