Agent Engineer

Zentis AI; Thiruvananthapuram; Sur place; INR 350 000 - 750 000; Plein temps

Résumé du poste

Zentis AI is seeking an Agent Engineer to design, build, and harden AI agents in production. You will own the full agent lifecycle from skills and tool integrations to guardrails and escalation logic, ensuring changes are safe to ship.

Work at the intersection of deterministic systems and language models, writing deterministic rules and wiring tools to systems of record. This senior IC role requires owning how a class of work gets done and defending that decision in writing to the team.

Qualifications

• 3+ years in software/ML engineering with strong Python proficiency.
• Hands-on experience with LLM frameworks (LangChain, LlamaIndex, LangGraph, CrewAI, or similar).
• Experience building with agentic patterns — tool use, ReAct, planning, memory.
• Solid understanding of prompt engineering, RAG, and function calling.
• Familiarity with LLM evaluation (RAGAS, DeepEval, custom harnesses).
• Experience with vector databases (Pinecone, Weaviate, Qdrant, pgvector).
• Strong grasp of API design, async programming, and distributed systems.

Responsabilités

• Design and implement autonomous/semi-autonomous AI agents with well-defined skills and tools.
• Build guardrails, safety layers, and escalation rules for high-stakes workflows.
• Develop and maintain evaluation frameworks (offline + online) for agent behavior.
• Optimize agent reliability, latency, and cost across production workloads.
• Collaborate with ML, Platform, and Solutions teams to ship agent capabilities end-to-end.
• Debug complex multi-step agent failures and continuously improve robustness.

Connaissances

Python

Software/ML engineering

LLM frameworks

Agentic patterns

Prompt engineering

LLM evaluation

API design

Distributed systems

Async programming

Outils

LangChain

LlamaIndex

LangGraph

CrewAI

Pinecone

Weaviate

Qdrant

pgvector

Description du poste

Build and harden agents in the stable: skills, tools, guardrails, escalation rules, and the evaluation that proves a change before it ships.

Objective

We're looking for an Agent Engineer to design, build, and harden AI agents in production. You'll own the full agent lifecycle — from skills and tool integrations to guardrails, escalation logic, and the evaluation harness that proves every change is safe to ship.

About the role

As an Agent Engineer at Zentis you'll design, build, and harden AI agents in production. You'll own the full agent lifecycle — from skills and tool integrations to guardrails, escalation logic, and the evaluation harness that proves every change is safe to ship.

You'll work at the intersection of deterministic systems and language models: most regulated BFSI work is narrow and well-specified, so your job is to make agents behave predictably — writing deterministic rules that run before any model does, wiring tools to systems of record, and turning "seems right" into a number the whole team can defend to a compliance audience.

This is a senior individual-contributor role with real ownership. You are not implementing tickets; you are deciding how a class of work gets done safely, and defending that decision in writing to the rest of the team.

Key responsibilities

• Design and implement autonomous/semi-autonomous AI agents with well-defined skills and tools
• Build guardrails, safety layers, and escalation rules for high-stakes workflows
• Develop and maintain evaluation frameworks (offline + online) for agent behavior
• Optimize agent reliability, latency, and cost across production workloads
• Collaborate with ML, Platform, and Solutions teams to ship agent capabilities end-to-end
• Debug complex multi-step agent failures and continuously improve robustness

Required skills

• 3+ years in software/ML engineering with strong Python proficiency
• Hands-on experience with LLM frameworks (LangChain, LlamaIndex, LangGraph, CrewAI, or similar)
• Experience building with agentic patterns — tool use, ReAct, planning, memory
• Solid understanding of prompt engineering, RAG, and function calling
• Familiarity with LLM evaluation (RAGAS, DeepEval, custom harnesses)
• Experience with vector databases (Pinecone, Weaviate, Qdrant, pgvector)
• Strong grasp of API design, async programming, and distributed systems

Nice to have

• Experience with observability tools (LangSmith, Langfuse, Arize)
• Background in BFSI, compliance, or regulated environments
• Contributions to open-source agent frameworks

Skills & technologies

Tell us what you would build and why it matters. If the timing is right, the conversation goes from there.