AI Document Search for Mining Procedures
RAG-powered search across 4,000+ safety and procedure documents. Workers ask questions in plain English and get sourced answers in seconds.
Mining operator — safety, training and compliance departments
A large mining operation with over 4,000 safety procedures, work instructions, training materials and compliance documents accumulated over 15 years. Documents lived across SharePoint, network drives, and legacy document management systems.
Finding the right procedure meant knowing which folder it was in, which version was current, and which document applied to which site or equipment type. New staff had no chance. Even experienced supervisors often worked from memory rather than searching for the actual procedure.
What needed to change
Nobody could find anything. 4,000+ documents across multiple systems with inconsistent naming, no tagging and no central search. Workers defaulted to asking a colleague or working from memory — which created safety and compliance risk.
Version control was broken. Multiple versions of the same procedure existed across different folders. There was no reliable way to know if the document you found was the current version or something superseded 3 years ago.
Search had to be dead simple. The end users were miners, operators and field supervisors — not office workers. Any solution that required specific keywords, boolean operators, or knowing which system to search was going to fail.
What we built
An AI-powered knowledge search system using RAG architecture. Workers ask questions in plain English and get answers with direct citations to the source documents.
Document Ingestion Pipeline
Automated ingestion from SharePoint, network drives and legacy systems. PDFs, Word docs and scanned documents are processed, chunked and embedded for semantic search.
RAG Search Engine
Retrieval-Augmented Generation search — workers ask a question in plain English and the system retrieves relevant document chunks, then generates a clear answer with source citations.
Source Citations
Every answer includes links to the specific source documents and sections. Workers can click through to read the full procedure. Builds trust in the AI-generated responses.
Role-Based Access
Search results respect existing document permissions. Workers only see documents they have access to based on their role, site and clearance level.
How it works
Worker asks a question
Types a natural language question into the search bar — e.g., "What is the isolation procedure for the primary crusher at Site 3?"
System retrieves relevant chunks
The RAG engine searches the vector database for the most relevant document sections based on semantic similarity to the question.
AI generates an answer
GPT-4 generates a clear, concise answer based only on the retrieved document content. No hallucination — the model is constrained to source material.
Citations displayed with answer
The answer includes clickable citations to specific documents and sections. The worker can verify the answer against the original source.
Feedback loop improves results
Workers can rate answers and flag incorrect responses. Feedback is used to refine chunking, embedding and retrieval strategies over time.
Measurable outcomes
Blokes on site can now find the right procedure in 30 seconds instead of 15 minutes — or instead of just guessing. That is a genuine safety improvement, not just an efficiency gain.
How we delivered it
Document Audit
2 weeksCatalogued all document sources, formats, naming conventions and access controls. Identified 4,200+ documents across 3 systems, with significant duplication and version conflicts.
Ingestion & Embedding
4 weeksBuilt the document processing pipeline — PDF extraction, OCR for scanned docs, chunking strategy, embedding generation. Resolved version conflicts and established the canonical document set.
RAG Engine & UI
4 weeksBuilt the retrieval and generation pipeline with GPT-4. Developed the search interface optimised for simplicity. Implemented citation linking and role-based access controls.
Testing & Refinement
3 weeksTested with real queries from safety managers and field supervisors. Refined chunking strategy and retrieval parameters based on answer quality. Validated access controls across all roles.
Similar Project?
Want something similar for your business?
Tell us about your industry, your workflows, and what you want to achieve. We will scope it, quote it fixed-price, and build it.
Tell Us About Your Project
What industry are you in? What systems do you use? What is the biggest operational problem you want solved? We will come back with a plan and fixed-price quote.
Prefer a quick chat? Call 0425 531 127 – we're Perth-based and we answer the phone.