Agentic AI Engineer

As an Agentic AI Engineer at Thinqmesh, you will build software that does more than answer. Your agents will plan a task, call the right tools, recover from failure and hand over to a person when a decision needs one. You will take agents from an exciting prototype to something an organisation can rely on every day.

We already expose our own systems to agents through MCP, so you will be working with this way of building as part of everyday practice rather than as an experiment.

Agents are the area of AI that organisations are moving on fastest, and agentic skills are among the most quickly growing in the job market. Employers increasingly ask for proof that an agent has been made safe and measurable, with least-privilege permissions and proper evaluation, because that is what stands between a clever demo and a system trusted with real work.

You will design agent workflows and multi-step orchestration, and you will build secure tool integrations using function calling, structured outputs, APIs and MCP. You will decide where an agent should act alone and where a person should approve the next step.

You will engineer for what goes wrong, adding validation, least-privilege permissions, retries, timeouts and rate limits. You will build evaluations for complete agent runs and trace behaviour in production, so that every action can be understood and improved.

You bring at least three years of software engineering, with hands-on experience building agents using LangGraph or a similar framework. You know MCP and tool calling well, you can evaluate an agent rather than just run it, and you think carefully about security and cost.

Experience with multi-agent patterns, A2A, sandboxed execution or observability tools would be a real advantage, as would examples of agents you have shipped to real users.

This role is hybrid or remote. It is full time in Noida, and a consultant role in San Francisco, Tustin and London. You will work closely with the client teams and colleagues you support.

We read every application ourselves. Show us something you have built or shipped, because real work gives us far more to talk about than a list of tools.