Lead Autonomous AI Engineer for Production Agentic Systems

EPAM Systems; Argentina; On-site; ARS 90,953,000 - 166,748,000; Full time

Benefits offered by this job

Healthcare benefits

Paid time off

Training and certifications

Global career opportunities

Volunteer opportunities

Job summary

EPAM Systems is seeking a Lead Agentic AI Engineer to design, build, and operate scalable autonomous agent systems for production AI deployments. You will drive decisions across runtime, memory, identity, tools, and observability, while leading adoption of AWS Bedrock AgentCore and related SDKs.

You will architect end-to-end systems, lead SDK implementations, and design memory-management capabilities to support sustained multi-step engagements.

Qualifications

• 5+ years building production software.
• 2+ years deploying LLM-based agents to production.
• Proficient in Python and agentic frameworks.
• Experience with AWS Bedrock AgentCore components.
• Strong understanding of observability, memory management, and security boundaries.

Responsibilities

• Architect end-to-end autonomous agent systems for production use.
• Lead adoption of AWS Bedrock AgentCore and related SDKs.
• Design long-term memory and context management for multi-step engagements.
• Build and maintain MCP servers and agent tool integrations.
• Optimize agent reasoning for reliability and cost-efficiency.
• Define standards for evaluations, tracing, and guardrails.

Skills

Python

Agent frameworks

LLM agents

Observability

Memory architectures

Prompt engineering

Security boundaries

Tools

Strands

LangGraph

Claude Agent SDK

OpenAI Agents SDK

CrewAI

AWS Bedrock AgentCore

Job description

EPAM Systems is seeking a Lead Agentic AI Engineer to design, build, and operate scalable autonomous agent systems for production AI deployments. You will drive decisions across runtime, memory, identity, tools, and observability, while leading adoption of AWS Bedrock AgentCore and related SDKs.

You will architect end-to-end systems, lead SDK implementations, and design memory-management capabilities to support sustained multi-step engagements.