Agent engineer

Role: Agent Engineer

Experience: 3 - 6 years

Location: Bengaluru

Role Summary

We are building an AI-native B2 B intelligence platform. You will build our core Agentic Harness — a performant, configurable agent layer that powers everything from our database search to sourcing and selling agents. This is a hands-on builder role focused on system design, observability, and agent performance.

Roles & Responsibiliti

Build a lightweight, scalable agent harness that allows the business to configure and deploy multiple domain-specific agents. Design the core orchestration loop, including memory management, state handling, tool calling, and deterministic guardrails. Ensure the agent layer is model-agnostic, easily swappable, and testable across different commercial and open-source LLMs. Optimize agent performance for low latency, efficient token usage, and minimal memory footprint. Establish robust evaluation frameworks (evals) to measure agent reasoning, tool selection accuracy, and task completion. Collaborate with the research team to tune open-source LLMs specifically for optimal agentic behavior and tool use.

Skills Required

Must Have:

3+ years of experience in backend or ML engineering, with strong Python skills. Experience building agentic workflows or orchestration loops Deep understanding of LLM application architecture, state machines, and API integrations (Fast API, Postgre SQL). Strong systems engineering mindset—you treat agent development as reliable software engineering, not just prompting. Experience benchmarking, profiling, and optimizing LLM performance and latency.

Good to Have:

3+ years of experience in backend or ML engineering, with strong Python skills. Experience building agentic workflows or orchestration loops Deep understanding of LLM application architecture, state machines, and API integrations (Fast API, Postgre SQL). Strong systems engineering mindset—you treat agent development as reliable software engineering, not just prompting. Experience benchmarking, profiling, and optimizing LLM performance and latency.