RAG — Technical Document Search
Overview
PlantPulse RAG is a RAG (Retrieval-Augmented Generation) search server that lets AI instantly search and reference technical documents, manuals, SOPs, and regulations from the manufacturing field.
It provides accurate, context-aware document search through four search modes, and also supports multimodal (image, table, etc.) search.
Document search is split into two layers.
- RAG API (
plantpulse-rag-api) — the layer that users and AI actually interact with. It provides the document management screens and the search API. - LightRAG — the internal engine that performs the actual indexing and search. It is not exposed externally and is reached only through the RAG API.
In other words, there is no way to open LightRAG's address or WebUI (/webui) directly. All document
work is done from the RAG API screens. (The legacy in-house server plantpulse-rag has been archived, and
the direct-upload path through the LightRAG WebUI was replaced by the Documents screen below as of
2026-07-30.)
Search Modes
| Mode | Description | Best suited for |
|---|---|---|
| hybrid (default) | Dense + Sparse combined | General document search (recommended) |
| naive | Semantic (meaning-based) search | Conceptual questions, finding similar documents |
| local | Detailed search within a document | Checking specific details of a particular document |
| global | Search across document relationships | Consolidated information spanning multiple documents |
Access Paths
| Task | Where |
|---|---|
| Ask via chat | AI Chat Web — the built-in agent calls the rag_search tool automatically |
| Run a search directly | Query screen of the RAG API dashboard |
| Upload, delete, or list documents | Documents screen of the RAG API dashboard |
| Check auto-sync results | Sync History screen of the RAG API dashboard |
| Investigate indexing failures | Logs screen of the RAG API dashboard |
Two Paths for Incoming Documents
Documents are indexed in one of two ways. How you manage them depends on which path they came in through.
| Path | Source | Description |
|---|---|---|
| Auto sync | Documents attached to equipment in Platform | When you attach a document to equipment in Platform, it is indexed automatically on a schedule. There is no need to upload it separately on the RAG side |
| Manual upload | Documents screen | Upload documents that do not belong to any equipment (common regulations, training materials, etc.) directly. The originals are stored separately and survive re-indexing |
Manually uploading a document that belongs to equipment into RAG will put it out of sync with Platform's document ledger. Management stays simpler if you attach equipment documents to the equipment in Platform and leave RAG to auto sync.
Multimodal Support
Search over multimodal content — text plus images, tables, and more — is supported:
- Attach an equipment photo and ask "What is this part?"
- Photo of an instrument panel → search the related configuration manual
- Image URL or Base64 data supported
From Registration to Searchable
문서 등록 (플랫폼 설비 첨부 / Documents 화면 업로드)
│
▼
파싱 — PDF·DOCX·이미지에서 텍스트와 표를 추출
│
▼
청킹 — 검색 단위로 잘게 나눔
│
▼
임베딩 — 각 조각을 벡터로 변환
│
├──▶ 벡터 저장소 (유사도 검색)
└──▶ 지식 그래프 (엔티티·관계 추출)
A document is not searchable immediately after registration. It appears in results only after parsing and embedding are complete, so if a document you just uploaded is not being found, check its processing status on the Sync History screen first.
Manageable Document Types
| Document type | Examples | AI use |
|---|---|---|
| Maintenance manuals | Periodic inspection procedures per equipment, lubricant replacement intervals | Provides appropriate countermeasures immediately when an anomaly is detected |
| SOPs | Work procedures per process, safety procedures, quality inspection criteria | Accurate step-by-step guidance |
| Safety regulations | Chemical MSDS, occupational health and safety rules, emergency response manuals | Immediate reference to regulations in hazardous situations |
| Technical specifications | Equipment specs, sensor specifications, allowable ranges | Reference for anomaly judgment thresholds |
| Incident/failure reports | Past failure cases, root cause analyses, recurrence prevention measures | Cause estimation based on similar cases |
| Training materials | New-hire training texts, process descriptions, quality training | Support for field training |
Configuration
Search parameters such as the embedding model, chunk size, and Top-K are LightRAG engine settings and cannot be changed from the UI. The values currently in effect can be viewed on the Config screen of the RAG API dashboard (tokens and passwords are shown masked).
What you typically need to adjust during operation is on the sync side.
| Item | Description |
|---|---|
| Sync enabled | Turns automatic indexing of Platform equipment documents on and off |
| Sync interval | How often Platform is checked for document changes |
| Documents indexed per run | Upper limit processed per cycle — used to control load during a large initial influx |
| Maximum file size | Documents larger than this are skipped |
Usage Scenarios
| Scenario | Search example |
|---|---|
| Check a maintenance procedure | "How to replace the injector bearing" → maintenance manual |
| Reference a safety regulation | "Response procedure for a nitrogen gas leak" → emergency response manual |
| Anomaly response guidance | "Inspection items for a temperature sensor fault" → inspection SOP |
| Search past cases | "Cause of last year's Line 3 injector failure" → failure report |
| Verify a specification | "Allowable temperature range for the injector" → technical specification |