Skip to content
Justina.
Back to work

Banking AI POC

SOP → Automated Journey Map Generator

AI / Full-stack Engineering · FastAPI · AWS Bedrock · RAG

Year
2026
Engagement
Datamellon
Focus
AI, RAG, Backend

I built an AI-powered proof of concept that transforms standard operating procedure (SOP) documents into branded, production-ready journey maps.

The system combines document processing, retrieval-augmented generation, LLM reasoning, and programmatic image rendering into a single automated workflow.

What I Built

The application accepts SOP documents and processes their contents to generate structured journey-map information before rendering the result as a branded PNG.

Workflow

  1. SOP Document
  2. Document Parsing
  3. Knowledge Retrieval
  4. LLM Processing
  5. Journey Map Structure
  6. Branded PNG Output

Technical Implementation

I developed the service using FastAPI and Uvicorn, with AWS services powering the AI and retrieval layer.

AI & Retrieval

  • Amazon Nova Pro through AWS Bedrock
  • AWS Bedrock Knowledge Base
  • Hybrid retrieval combining vector and BM25 search

Document Processing

  • pypdf for PDF processing
  • python-docx for Word documents

Rendering

  • Pillow for programmatic journey-map generation
  • Custom rendering logic for structured layouts
  • Client branding incorporated directly into the generated output

The Problem I Solved

Creating journey maps from SOP documents can involve manually reading, interpreting, structuring, and visually representing large amounts of information.

I built the POC to demonstrate how this process could be automated using AI while still producing a structured and visually consistent output.

Engineering Highlights

This project gave me hands-on experience combining traditional backend development with modern AI infrastructure — particularly document processing, RAG, LLM integration, and programmatic visual generation.