Aws Agentic Document Assistant Upgraded

Agentic Documents Assistant — Upgraded for Current AWS Tooling

An LLM assistant that answers complex questions from business documents through natural conversation — combining semantic search (RAG), SQL generation over extracted entities, and an LLM agent that decides which approach a given question needs.

This is a **modified, upgraded fork** of AWS's [`aws-agentic-document-assistant`](https://github.com/aws-samples/aws-agentic-document-assistant) sample repository, updated to deploy successfully against **current Bedrock models, current LangChain APIs, and current CDK/Node/SageMaker tooling** — none of which the original sample runs against unmodified anymore.

Acknowledgement

This project is a modified version of AWS's [Agentic Documents Assistant](https://github.com/aws-samples/aws-agentic-document-assistant), originally designed by Mohamed Ali Jamaoui, with contributions from Giuseppe Hannen and Laurens ten Cate. The original repository is licensed under MIT-0. All credit for the underlying architecture — the RAG + SQL + ReAct agent design — belongs to the original authors; this fork focuses on bringing the deployment and integration layer forward to current tooling.

What's different from the original

The original repo's `README` and code assume tooling versions from when it was published. Deploying it unmodified today fails at multiple points. This fork fixes:

• CDK/Node tooling drift — Aurora PostgreSQL engine version conflicts, CDK CLI/library version mismatches, Amplify's Node 18 deprecation
• Bedrock model lifecycle — the original pins anthropic.claude-v2, which (along with claude-instant-v1 and early claude-3-haiku) has since been retired or marked legacy by AWS
• The completion-to-chat API migration — Claude 3+ models require LangChain's BedrockChat class and the Messages API, not the older completion-style Bedrock class the original code uses
• Cross-region inference profiles — current high-demand models require i