Agentic AI Engineer

EY; Hyderabad, Bengaluru, Delhi; On-site; INR 6,000,000 - 9,000,000; Full time

Job summary

EY is seeking an experienced engineer to design, build and operate production-grade agentic AI systems in Python. You will implement end-to-end agentic workflows with planning, tool usage, multi-agent coordination, retrieval, evaluation, safety controls and observability.

You will work with LangGraph and Semantic Kernel, ensure robust tool execution, and optimize RAG pipelines in a Cloud environment. Strong engineering fundamentals and production delivery experience are required.

Qualifications

• Strong Python engineering: async patterns, modular code, pytest.
• Hands-on experience delivering production agentic systems.
• Deep understanding of RAG end-to-end: ingestion, retrieval and grounding.
• Multi-agent patterns: coordinator/worker, planner/executor.
• Experience with evaluation pipelines and model quality checks.

Responsibilities

• Build agentic workflows in Python using LangGraph and/or Semantic Kernel.
• Implement robust tool calls with validation, retries, and safeguards.
• Develop and optimize RAG/Agentic pipelines including embedding and hybrid retrieval.
• Design multi-agent coordination patterns and task decomposition.
• Drive observability, monitoring, and release readiness.

Skills

Python engineering

Distributed systems

API design

Security best practices

Evaluation pipelines

Education

Bachelor's degree in Computer Science or related field

Tools

LangGraph

Semantic Kernel

LangChain

AutoGen

CrevAI

Docker

Job description

Design, build, and operate production-grade agentic AI systems in Python using modern agent frameworks. You will implement complete agentic workflows including planning, tool use, multi-agent coordination, retrieval (RAG), evaluation, safety controls, and observability. The role requires strong engineering fundamentals, hands‑on experience shipping agentic systems, and the ability to continuously improve reliability and quality using evaluation-driven iteration.

Responsibilities-

• Build and operate agentic workflows in Python using LangGraph and/or Semantic Kernel (LangChain/AutoGen/CrevAI as additional exposure).
• Implement robust tool/function calling: schema validation, structured arguments/outputs, safe tool execution, retries, idempotency patterns, and permission‑scoped tool access.
• Develop and optimize RAG / Agentic RAG pipelines: ingestion, chunking, embeddings, hybrid retrieval, reranking, query planning, grounding, caching, and citation/trace strategies.
• Design multi‑agent systems: coordinator‑worker, planner‑executor, task decomposition, delegation, and optional human‑in‑the‑loop checkpoints.
• Drive evaluation and quality systems: regression suites, grounding/faithfulness checks, hallucination detection, task success metrics, latency/cost monitoring, and release gates.
• Use cloud AI platform services (AWS preferred) for model access/orchestration and embeddings, and integrate these into enterprise‑grade agent systems.
• Ensure production readiness: observability (logs/metrics/traces), dashboards and alerting, and operational runbooks for common failure modes.
• Apply security best practices for LLM/agentic systems: secrets handling, least privilege, prompt‑injection defenses, content safeguards, audit logging, and safe integration boundaries.
• Collaborate with architects, backend/platform engineers, and product teams to improve reliability, performance, cost, and UX.

Must Have-

• Strong Python engineering: async patterns, clean modular code, testing (pytest), profiling/performance debugging.
• Hands‑on experience delivering production agentic systems using LangGraph and/or Semantic Kernel, including tool orchestration and multi‑step workflows.
• Deep understanding of RAG systems end‑to‑end: ingestion retrieval grounding, including hybrid search and retrieval optimization concepts.
• Multi‑agent patterns and protocols: coordinator‑worker, planner‑executor, task decomposition; awareness of MCP and A2A‑style interoperability patterns.
• Evaluation and quality depth: experience with promptfoo, Phoenix/Arize (or equivalent), custom eval pipelines.

Cloud Platform AI Services Experience (AWS Preferred)

• Using managed model/embedding services (e.g., AWS Bedrock or equivalents) and integrating them into production systems.

Practical Production Engineering Fundamentals

• API design, auth/authz concepts, rate limiting, retries, idempotency, and resilient service patterns.

Experience Deploying Services

• Using Docker and CI/CD pipelines; ability to work with cloud runtime environments.

Security Awareness for GenAI

• Prompt injection defense, safe tool execution boundaries, secrets management, audit logging.

Good to Have-

• AWS infrastructure depth: EKS/ECS, S3, RDS, ElastiCache (Redis), SQS/SNS, API Gateway, OpenSearch / OpenSearch Serverless, Secrets Manager.
• Observability tooling depth: Datadog, OpenTelemetry, CloudWatch, distributed tracing, SLO‑style alerting.
• Enterprise integrations exposure: SAP / Salesforce / ServiceNow (or similar), API governance, throttling patterns.
• Streaming/chat‑based UX patterns: Trace visibility (server‑sent events, token streaming).
• Strong system design: Scalable architectures, failure mode analysis, cost/performance tradeoffs.