For the complete machine-readable documentation index, see llms.txt.
← Back to projects

2025-03-15

AWS Document Agent

Autonomous agent that assembles four classes of regulatory documents end-to-end using Bedrock, Step Functions, and guardrails.

AgentsAutomationServerless

Outcome

Removed 95% of manual effort for dynamic document creation.

Stack

AWS BedrockLangChainStep FunctionsDynamoDBEventBridge

Related service

AI Agents

Custom internal agents for document operations, workflow execution, approvals, and controlled human-in-the-loop systems.

Explore AI Agents →

Scope

Legal and operations teams generated four templates manually: onboarding packs, SLA annexes, ESG addenda, and vendor scorecards. Each document pulled from different APIs, spreadsheets, and knowledge bases.

Architecture

  • LangChain orchestrates tool-aware agents powered by Bedrock Claude.
  • Retrieval layer mixes Bedrock Knowledge Bases with S3 semantic stores for contextual grounding.
  • Step Functions manages validation, human approvals, and retries, persisting state in DynamoDB and emitting events to EventBridge webhooks.

Impact

Documents now assemble in under two minutes with structured citations. FinOps dashboards track token usage per template, and guardrails enforce policy prompts before sending PDFs to clients.

Projects

Need this kind of system in your team?

I help teams ship document agents, RAG copilots, computer vision pipelines, and operational automations without the usual prototype-to-production gap.

Start a project conversationivanncaamano@gmail.com