Overview
I built a retrieval-augmented assistant to help production and maintenance teams quickly answer process questions from internal documentation.
Problem
Important operating knowledge was spread across SOP PDFs, operator notes, and legacy manuals. Finding the right answer often took too long during active line issues.
What I built
- Ingestion pipelines for PDF and markdown process documents
- Chunking and embedding workflow with metadata for plant area and machine family
- Retrieval and answer generation flow with source citations
- Guardrails for low-confidence responses
Outcome
The assistant reduced time-to-answer during troubleshooting and made knowledge transfer easier between shifts.
Next improvements
- Add role-aware access controls by site and department
- Add feedback capture to improve retrieval quality over time
- Extend to maintenance logs for better fault-history grounding