Associate Director - Agentic AI

Description :

Location : Bengaluru (BFSI, Product & Tech firms)

Exp : 7 -16 yrs

Must-Have (Non-Negotiable) :

- Python Coding hands-on

1. Agentic AI :

- Deep, Hands-On Experience ( Google ADK/AutoGen/LangGraph- more the better)

- Agentic AI (not just LLM wrappers).

- Google ADK (Agent Development Kit) - explicit requirement.

- LangGraph (explicit).

- AutoGen (Microsoft) explicit.

- Multi-Agent Frameworks - building, orchestrating, supervising multiple agents.

Architecture :

From Scratch & Enterprise :

- End-to-end architecture for Agentic AI systems.

- Enterprise architecture : Ability to design at scale, not just POCs.

- Architecture from scratch : Greenfield / zero-to-one capability.

- MLOps for Generative AI / Agents

- MLOps specifically applied to LLMs, agents, and multi-agent systems.

- Model deployment, monitoring, versioning, and retraining pipelines for agentic workloads.

Docker & Containerization :

- Creating and owning Docker images (not just using pre-built ones).

- Deep understanding of containerization for agentic AI services.

Financial Services & Scale :

- 13+ years (as in original JD) in Analytics / Data Science within Financial Services.

- Proven track record of taking agentic AI to production in regulated environments.

Programming & Core AI :

- Python (advanced, hands-on).

- LLM fundamentals - prompt engineering (Chain-of-Thought), RAG, vector databases.

- Schema validation, structured outputs, reflection patterns.

Good-to-Have (Strongly Preferred / Differentiators) :

These separate qualified from the one they hire.

Additional Agentic Frameworks :

- CrewAI

- Semantic Kernel

- LangSmith (for tracing & observability)

Advanced MLOps for Agents :

- MLflow, Kubeflow, or Vertex AI Pipelines (especially on Google Cloud, given ADK).

- Agent evaluation frameworks (e.g., AgentBench, custom evals).

- LLM observability tools (LangFuse, Arize, Weights & Biases).

Cloud & Infrastructure :

- Google Cloud Platform (GCP) : Strongly implied by Google ADK.

- Kubernetes (K8s) : Deploying agent containers.

- Terraform / Infrastructure as Code.

- Operational Memory & State Management

- Vector databases (Pinecone, Weaviate, Milvus, Chroma).

- Semantic caching for agents.

- Long-term memory patterns for multi-agent systems.

- Traditional Analytics / DS (from original JD)

- A/B testing frameworks for agentic systems.

- NLP / deep learning / reinforcement learning (nice to have, not core).

- Governance in Regulated AI

- Experience with model risk management for LLMs/agents.

- Audit trails for agent decision-making