NSP AI Enquiry Workflow

This project delivers a practical AI-driven intake workflow for NSP Cases. It reads customer enquiry emails for custom flight cases, extracts technical/commercial details with an LLM, and outputs a clean JSON payload ready for downstream operations.

After the initial version, the system was upgraded to a service-style architecture with:

FastAPI backend endpoints
browser-based local UI
Docker and Compose runtime support
test and evaluation scaffolding for reliability

The design remains local, practical, and interview-ready:

core extraction logic isolated in main.py
prompt files externalized for fast iteration
provider abstraction for future model/provider swaps
stable normalized JSON schema for ERP/MRP integration

Repository

nsp-ai-enquiry-workflow

What Was Implemented

Input ingestion from sample_email.txt
Prompt-driven extraction of:
- product type
- dimensions (length, width, height, unit)
- use case
- requirements
- attachment mention detection
- concise business summary
- missing information list
- confidence score
JSON normalization and validation handling in Python
Structured output writing to output/example_output.json
Human-review-ready pattern via missing_information and confidence
Added FastAPI service layer in app.py with:
- POST /api/extract
- GET /health
- GET /version
Added lightweight frontend (templates/ + static/) for browser-based extraction usage
Added containerized runtime support:
- Dockerfile
- docker-compose.yml
Added quality and reliability scaffolding:
- API/unit tests in tests/
- evaluation starter in evaluation/ (offline and live modes)