Enterprise Agentic AI Platform

The problem

Data science teams wanted to use LLM agents for knowledge-heavy work: finding answers across internal documents, supporting decisions, and moving work through multi-step processes. Each team was starting from scratch, and getting a prototype to production meant solving the same infrastructure problems again and again.

What I built

I architected and deployed a shared agentic AI platform so teams could focus on their use case instead of the plumbing.

• Agent runtime built on the AWS Strands framework, with LLM tool-calling so agents can query systems and act, not only chat.
• Retrieval layer using RAG and vector search, so agents answer from the organization’s own knowledge.
• Model hosting on Amazon SageMaker for enterprise deployment.
• Hierarchical workflows, where agents coordinate across the multi-level processes common in banking.

Key decisions

• Platform, not projects. Investing in reusable SDK components (see Agent SDK & observability) meant each new use case got cheaper to deliver.
• Evaluation from day one. Every use case ships with checks on retrieval quality, response quality, and agent behavior. See GenAI evaluation.

My role

Forward-deployed engineer embedded with the client’s teams. I owned use cases end to end through the AI development lifecycle: scoping, design, build, evaluation, and production deployment.