Skip to content

Onboarding Assistant

AI customer onboarding and policy recommendation, from identity check to submitted application

| Documentation | Live demo | Agent console | Architecture | Demo walkthrough | Development |

About

An assistant that takes an insurance customer through identity verification, customer profiling, policy recommendation and policy application, while support agents watch every session and take over when needed.

  • A LangGraph graph runs the four stages. It pauses whenever it needs a person and resumes from an encrypted Postgres checkpoint; failures and dead ends hand the session to an agent.
  • Code decides, the LLM explains. Eligibility, ranking and price are deterministic code; Claude Sonnet 4.6 on Amazon Bedrock extracts values from free text and writes explanations and summaries.
  • Two services on AWS. A Next.js frontend (customer app and agent console) and a FastAPI + LangGraph backend on ECS Fargate, provisioned with Terraform and deployed with GitHub Actions.
  • Runs anywhere without AWS. One mock service stands in for the partner, identity, contract admin systems and Bedrock, with four seed customers that each show a different path.
Two services, one database, four external systemsCustomer/s/{token}Support agent/agentFrontend service: Next.jsCustomer app+ agent console/api/* relay(incl. SSE)Backend service: FastAPI + LangGraphOnboarding graphcode / LLM / wait nodesPostgreSQLcheckpoint · domain · catalogExternal systems (mock locally)PartnerIdentity (OTP, document)Contract adminBedrock: Claude Sonnet 4.6Built in this repoExternal system, mocked locally

Getting started

docker compose up --build

Open http://localhost:13000 and start as a customer; the agent console is at http://localhost:13000/agent. The demo walkthrough takes you through the four seed customers; the development guide covers the other services, the repository layout and the tests.

Documentation

The design documents live in docs/ and are published at dev.docs.onboardassist.click:

  • Design: solution architecture, LangGraph design, state management, data model, observability
  • Infrastructure: AWS architecture, networking, Terraform, CI/CD
  • Decisions: assumptions, tradeoffs, and future improvements with the known limits
  • Research: the notes behind the design

docs/README.md maps each document the brief asks for to the section that answers it. The contracts between the services are in CONTRACTS.md.