AI · ongoing project
An AI chain being put in place
On an ongoing project: an AI chain with LangGraph, Bedrock and Pydantic, on a serverless AWS infrastructure with a GraphQL API. The model and the storage are replaceable components.
Context
An ongoing project, whose client and nature are not described here. What can be: the way the chain is built to stay reversible, and the tools chosen.
Orchestration in the code, not at a provider
LangGraph describes the flow as a graph: steps, conditions, pauses for human validation. The library is free software hosted in the project’s infrastructure; state is stored in a database the client controls.
The model behind an interface
Models are called through Bedrock, behind a single interface. Responses are validated by Pydantic schemas: anything that does not match the expected format is rejected before reaching the rest of the chain. Changing model or provider does not touch the application.
Serverless, GraphQL, reversible
Serverless infrastructure on AWS, GraphQL API with server-side authorization, separate environments. Business rules stay in the client’s repository; the model and the storage are replaceable components.
Result
Frequently asked questions
Why LangGraph?
Because it describes the flow as a readable graph, handles state and pauses for human validation, and remains a free library you host yourself.
Why Pydantic?
So that every model response is validated against a schema before being used. A model returns text; Pydantic turns it into reliable data or an explicit error.
Does the project depend on AWS?
The infrastructure is on AWS, but the orchestration, the schemas and the business rules are in the client’s code, and model access goes through a replaceable interface.
Related services
A comparable case on your side?
Tell me which process costs you time; I get back within 24 business hours with a first read, no obligation.