RAG semantic intelligence and document search
Retrieval-Augmented Generation, vector embeddings, natural language semantic search, and conversational document interaction.
1. Overview & Business Value#
This comprehensive operational guide details the architecture, business workflows, configuration parameters, and execution steps for RAG semantic intelligence and document search within SmartERP.

Beyond keyword queries, SmartERP Documents Enterprise incorporates Retrieval-Augmented Generation (RAG) powered by Gemini AI and vector embeddings. This allows users to converse directly with their organization's knowledge base and retrieve answers synthesized across hundreds of documents.
2. Prerequisites & Security Access Groups#
- Required Modules:
documents, smarterp_dms - Security Groups:
Documents User(Technical ID:documents.group_documents_user)Documents Administrator(Technical ID:documents.group_documents_manager)
- Master Data Dependencies: Workspace hierarchies, document tags, automated actions, OCR rule definitions, and retention policies.
3. Step-by-Step Configuration & Execution#
Step 1: Navigating to the Interface#
- Log into your SmartERP instance at
https://smarterp.et/odoo. - Open the main App Switcher (grid icon in the top-left corner).
- Navigate to Settings → Technical.
Step 2: Operational Data Parameters & Controls#
Ensure all mandatory parameters comply with enterprise operational standards:
- Authorized Operator: Action executed under assigned RBAC permissions (System Administrator, Certified Specialist).
- Relational Integrity: Linked to upstream master records, approval requests, or transactional ledgers.
- Statutory & Quality Controls: Verified against Ethiopian statutory regulations and corporate governance standards.
Step 3: Operational Execution & State Progression#
- State Progression Gate: Advance record from Draft through Audited to Confirmed / Certified.
- Integrated System Impact: Automatically updates real-time operational states and posts financial/inventory journal moves.
- Audit Verification: System registers tamper-evident audit records logging user ID, timestamp, and cryptographic transaction digests.
How RAG document intelligence works#
flowchart TD
A[Document Uploaded] --> B[Text Chunking & Tokenization]
B --> C[Gemini Embedding API]
C --> D[(PostgreSQL pgvector / Vector Index)]
E[User Asks: 'What are the delivery penalties in the 2026 Telecom contract?'] --> F[Query Embedding]
F --> G[Cosine Similarity Search against Vectors]
G --> H[Top 3 Relevant Contract Clauses Retrieved]
H --> I[Synthesized Answer with Exact Source Citations]
D -.-> G
Key capabilities#
- Natural Language Search: Find documents by meaning rather than exact keywords (e.g., searching "employee vehicle policy" retrieves documents titled "Automobile Fleet Operating Guidelines").
- Interactive Q&A: Ask questions about complex legal contracts or internal policy manuals and receive direct answers accompanied by page-number citations.
- Multilingual Queries: Inquire in Amharic about documents written in English, and receive synthesized answers in Amharic with translated context.
4. Advanced Configuration Points#
- Automated Actions & Scheduled Tasks (Crons):
Documents: Run OCR field extraction(documents.ir_cron_ocr_process): Runs every 15 minutes to process incoming scanned document textDocuments: Archive expired records(documents.ir_cron_retention): Runs weekly to enforce statutory document retention schedules
- Custom Print Templates & Reports:
- Document Audit Trail & Access Log
- Workspace Document Storage & Activity Ledger
- Multi-Company Rules: Record-level security isolation under Company 1 (
Ashewa Technology Solutions S.C.), preventing cross-company ledger leakage across parent and subsidiary legal entities.
5. Comprehensive Troubleshooting Matrix#
| Error Symptom / Exception | Probable Root Cause | Corrective Action / Resolution |
|---|---|---|
Access Rights Error: Workspace access restricted | User lacks permissions for confidential legal or HR workspace. | Request workspace access clearance from Documents Administrator. |
OCR Extraction Failure: Low scan resolution | Scanned image DPI is below minimum threshold for optical character recognition. | Upload document at minimum 300 DPI resolution with clear contrast. |
Upload Blocked: File size exceeds maximum limit | File exceeds system maximum attachment size. | Compress file or upload via secure external cloud link. |
Last updated: 2026-11-15