You will design and productionize agents that reason, act, and know when to stop and hand back to a person. The interesting work is not the demo — it is the boundary: which decisions an agent may close on its own, and which it must escalate with its reasoning attached.
What you’ll own
- Agent and tool design for live client workflows, from evaluation harness through to rollout.
- The escalation boundary, and the evidence that it is set in the right place.
- Latency, cost, and failure behaviour under real load — including what the agent does when a tool it depends on is down.
- The audit trail: every action an agent takes should be explainable after the fact, to a regulator if it comes to that.
What we look for
- Production experience with LLM applications — orchestration frameworks, retrieval, tool use, and evaluation you actually trust.
- Strong software engineering. Agents that survive contact with production are ordinary systems with an unusual component.
- Scepticism about autonomy for its own sake, and the ability to argue that position with data.
- Willingness to sit with the operators who will use what you build.
Where you’ll sit
Mumbai or Indore, with time on the client floor as engagements require. Tell us which base suits you when you apply.