I ship agentic AI to production. Then I teach your team to.

It is hands-on help designing and shipping AI systems that take actions, not just answer questions. In practice that means scoping the right workflow, building the agent with reliable tools, evals, and guardrails, and handing your team a system they can run. I work mostly with IT services teams putting their first or second agent into production.

A practical path from prompt to production agent: the agent loop, tool design, the Model Context Protocol, multi-agent orchestration, evals, and observability. Teams build real agents during the program rather than watching slides. Format and length are tailored to your team, from a focused two-day workshop to a multi-week cohort.

Whatever fits the job, but the stack I reach for most is the Claude API, the Model Context Protocol for tools, Python and FastAPI, pgvector for retrieval, and an observability layer like Langfuse or OpenTelemetry. The point is matching the tool to the problem, not standardising on a framework for its own sake.

For a bounded, well-scoped workflow, a working agent in front of real users in about 90 days is realistic, with the agent and its operating layer of evals, observability, and guardrails built together. What stretches timelines is rarely the model. It is unclear scope, messy data, and a security review that arrives late.

Yes. I am based in Ahmedabad and work with teams across India and remotely worldwide. Consulting and training are delivered in person or remotely depending on what suits your team.