Skip to content
Joshua Asoh

The Hartford · 2025 internship1 min read

A GenAI chatbot for internal documentation

A retrieval-augmented chatbot on Claude 3.5 Sonnet and AWS Bedrock that answers engineers' questions from internal documentation.

Role
Tech & Ops LDP Intern, Cloud and Reliability Engineer
When
May – August 2025
  • 1–2 hrsSaved per documentation lookup
  • 800+Engineers supported

The short version

Problem
Engineers spent one to two hours per question digging through unstructured internal documentation.
What I did
Led prompt engineering and designed a modular RAG pipeline on AWS Bedrock with Claude 3.5 Sonnet, plus an Angular chat UI.
Result
Lookups got one to two hours faster per query, and the chatbot scaled to 800+ engineers.

  1. Unstructured docs

    Internal documentation

  2. Processing pipeline

    Modular ingestion

  3. Retrieval

    RAG workflow

  4. Claude 3.5 Sonnet

    On AWS Bedrock

  5. Chat UI

    Responsive Angular app

My partContextRuns across AWS ECS, Lambda, and S3, provisioned with Terraform within IAM and security policy.

The problem

Answers buried in unstructured docs

Engineers were spending one to two hours tracking down answers spread across unstructured internal documentation.

Approach

Retrieval first, then generation

I led the prompt engineering and designed a modular pipeline that processes the unstructured documentation, retrieves the relevant passages, and has Claude 3.5 Sonnet on Bedrock answer from them.

  • Infrastructure provisioned with Terraform, compliant with AWS IAM and security protocols.
  • RAG workflow running across AWS ECS, Lambda, and S3.
  • A responsive Angular UI for real-time chat.

Outcome

Faster answers, at scale

The chatbot cut documentation lookup time by one to two hours per query and scaled to support more than 800 engineers.

Built with

  • AWS Bedrock
  • Claude 3.5 Sonnet
  • RAG
  • ECS
  • Lambda
  • S3
  • Terraform
  • Angular
Next case studyReal-time inventory for a research lab