Executive Snapshot

Audience: CXOs and founders running fine dining, catering, franchise groups, casual dining.
Core outcomes (what moves the business):
Cost savings: reduce waste and procurement errors, automate purchasing cycles with Craveva AI Enterprise.
Sales lift: increase AOV and conversion with Craveva AI Enterprise sales agents on web/WhatsApp/kiosks.
Time savings: remove manual exports, reporting, and SOP Q&A with Craveva AI Enterprise automation.
Operational consistency: standardize execution across outlets using Craveva AI Enterprise agents + data layer.

Platform Architecture (1 minute)

Data layer: connect POS, databases, Google Drive, and APIs into a unified view inside Craveva AI Enterprise.
AI layer: agents query and act on governed data (no fragile spreadsheet workflows) in Craveva AI Enterprise.
Deployment layer: deploy agents to WhatsApp, web widget, kiosks, or internal tools with Craveva AI Enterprise.

Execution Flow (Ops + Finance + IT)

Finance sets guardrails (approval thresholds, budgets, audit trail) in Craveva AI Enterprise.
IT connects data sources once; rollout scales outlet-by-outlet via Craveva AI Enterprise multi-outlet deployment.
Leadership tracks KPI movement weekly and expands successful automations with Craveva AI Enterprise.
Ops defines workflows (ordering, inventory alerts, SOP answers, customer responses) in Craveva AI Enterprise.

Setup Guide (fast path)

Connect data sources (POS + databases + Drive + APIs) in Craveva AI Enterprise.
Start with 2–3 agents: Procurement (cost), Sales (revenue), Analytics (visibility) in Craveva AI Enterprise.
Deploy to the workflow: WhatsApp/web/kiosk/internal portal using Craveva AI Enterprise.
Measure ROI and operational impact, then replicate across brands/outlets with Craveva AI Enterprise.

Leadership Metrics

Spoilage/expiry write-offs and transfer effectiveness
Stockout rate, lost sales signals, and substitution frequency
Procurement cycle time (draft ? approve ? receive)
Delivery basket value vs dine-in basket value (mix shift)
Peak-hour throughput (orders/hour) and queue time
Time-to-close (EOD) and reporting cycle time reduction

Next Steps

Documentation: /documentation
Models: /ai-models
Templates: /templates
Architecture: /solutions/architecture
Deployment: /solutions/deployment

RAG and Vector Databases: How Craveva AI Enterprise Processes Your Data

Learn how Craveva AI Enterprise uses RAG (Retrieval-Augmented Generation) and vector databases to process documents, files, and unstructured data, enabling AI agents to answer questions from your knowledge base with high accuracy.

Introduction

RAG (Retrieval-Augmented Generation) and vector databases are core technologies that enable Craveva AI Enterprise to process documents, PDFs, Word files, and other unstructured data. When you upload files to Google Drive or directly to the platform, RAG makes this content searchable and accessible to AI agents, enabling them to answer questions based on your actual documents.

What is RAG?

RAG (Retrieval-Augmented Generation) is a technique that combines:

Retrieval: Finds relevant information from your documents/data
Augmentation: Enhances AI context with retrieved information
Generation: Creates accurate, context-aware responses

Craveva AI Enterprise uses RAG to enable AI agents to answer questions from your documents, SOPs, training materials, and knowledge bases with high accuracy.

How RAG Works in Craveva AI Enterprise

flowchart TD
    subgraph Step1["STEP 1: DOCUMENT PROCESSING"]
        Upload[Upload Documents<br/>PDF, Word, Excel, Google Drive]
        Parse[Document Parsing<br/>Extract Text]
        Extract[Text Extraction]
        Chunk[Chunking<br/>Split into Smaller Chunks]
        Meta[Metadata Extraction<br/>Titles, Dates, Authors]
    end

    subgraph Step2["STEP 2: EMBEDDING GENERATION"]
        Embed[Create Embeddings<br/>Vector Representations]
        Model[Embedding Model<br/>Semantic Understanding]
        Vector[Vector Creation<br/>High-Dimensional Vectors]
    end

    subgraph Step3["STEP 3: VECTOR DATABASE"]
        Store[Store Embeddings]
        Index[Index for Fast Search]
        Scale[Scalable Storage<br/>Millions of Chunks]
    end

    subgraph Step4["STEP 4: QUERY PROCESSING"]
        Query[User Question]
        QueryEmbed[Query Embedding]
        Search[Similarity Search]
        Retrieve[Retrieve Relevant Chunks]
        Context[Add to AI Context]
        Generate[Generate Answer]
    end

    Upload --> Parse
    Parse --> Extract
    Extract --> Chunk
    Chunk --> Meta
    Meta --> Step2

    Step1 --> Embed
    Embed --> Model
    Model --> Vector
    Vector --> Step3

    Step2 --> Store
    Store --> Index
    Index --> Scale
    Scale --> Step4

    Query --> QueryEmbed
    QueryEmbed --> Search
    Search --> Retrieve
    Retrieve --> Context
    Context --> Generate

    style Step1 fill:#1e293b,stroke:#8b5cf6,stroke-width:2px
    style Step2 fill:#1e293b,stroke:#3b82f6,stroke-width:2px
    style Step3 fill:#1e293b,stroke:#10b981,stroke-width:2px
    style Step4 fill:#1e293b,stroke:#f59e0b,stroke-width:2px

Step 1: Document Processing

When you upload documents (PDF, Word, Excel, etc.) or connect Google Drive:

Document Parsing: Platform extracts text from documents
- PDF: Uses pdfjs-dist and pdf-parse for text extraction
- Word: Uses mammoth for .docx files
- Excel: Uses xlsx for spreadsheet data
- Google Docs: Direct text extraction via Google Drive API
Text Extraction: All text content extracted from documents
Chunking: Large documents split into smaller chunks (for better retrieval)
Metadata Extraction: Document titles, dates, authors extracted

Step 2: Embedding Generation

Craveva AI Enterprise creates embeddings (vector representations) for each chunk:

Text Chunks: Each document chunk converted to embedding
Embedding Model: Uses latest embedding models for semantic understanding
Vector Creation: Each chunk becomes a high-dimensional vector
Semantic Meaning: Vectors capture semantic meaning, not just keywords

Step 3: Vector Database Storage

Embeddings stored in vector database:

Vector Storage: All embeddings stored in optimized vector database
Indexing: Vectors indexed for fast similarity search
Metadata Storage: Document metadata stored with vectors
Scalability: Handles millions of document chunks efficiently

Step 4: Query Processing

When you ask a question:

Query Embedding: Your question converted to embedding
Similarity Search: Vector database finds most similar document chunks
Retrieval: Top relevant chunks retrieved (typically top 3-5)
Context Augmentation: Retrieved chunks added to AI context
Response Generation: AI generates answer using retrieved context

Vector Databases in Craveva AI Enterprise

Craveva AI Enterprise uses vector databases to:

Store Embeddings: Semantic representations of all document chunks
Enable Semantic Search: Find documents by meaning, not just keywords
Scale Efficiently: Handle large document collections (thousands of files)
Maintain Accuracy: Precise retrieval of relevant information
Fast Retrieval: Sub-second search across millions of chunks

Supported File Types

Craveva AI Enterprise processes:

PDF: Text extraction from PDF documents
Word: .docx files via mammoth
Excel: .xlsx, .xls files via xlsx
CSV: Comma-separated values
JSON: Structured data files
Text: Plain text files
Google Docs: Via Google Drive integration

RAG Implementation for Offline Files

Craveva AI Enterprise implements RAG for files uploaded directly or via Google Drive:

Local File Processing

flowchart TD
    Upload[1. Upload File<br/>PDF, Word, Excel, etc.] --> Process[2. Automatic Processing<br/>Extract Text]
    Process --> Chunk[3. Chunking<br/>Split into Manageable Chunks]
    Chunk --> Embed[4. Embedding<br/>Convert to Vectors]
    Embed --> Store[5. Storage<br/>Store in Vector Database]
    Store --> Ready[6. Ready for Queries<br/>Searchable by AI Agents]

    style Upload fill:#1e293b,stroke:#8b5cf6,stroke-width:2px
    style Process fill:#1e293b,stroke:#3b82f6,stroke-width:2px
    style Chunk fill:#1e293b,stroke:#10b981,stroke-width:2px
    style Embed fill:#1e293b,stroke:#f59e0b,stroke-width:2px
    style Store fill:#1e293b,stroke:#ef4444,stroke-width:2px
    style Ready fill:#1e293b,stroke:#10b981,stroke-width:3px

Upload File: Upload PDF, Word, Excel, etc. to platform
Automatic Processing: Platform extracts text automatically
Chunking: Large files split into manageable chunks
Embedding: Each chunk converted to vector
Storage: Vectors stored in vector database
Ready for Queries: File content now searchable by AI agents

Google Drive Integration

Connect Google Drive: Link Google Drive account (OAuth or public links)
Select Documents: Choose which Google Docs to process
Automatic Sync: Platform processes documents automatically
Continuous Updates: New documents processed as they're added
Searchable: All Google Docs content searchable by AI agents

Use Cases

Craveva AI Enterprise RAG powers:

Customer Service

FAQ Documents: AI Customer Service Agent answers from FAQ documents
Product Manuals: Answers questions about products from manuals
Policy Documents: Explains policies from company documents

Staff Training

SOP Documents: AI Internal Coach answers questions from SOP documents
Training Materials: Provides training guidance from materials
Employee Handbooks: Answers HR questions from handbooks

Data Analysis

Reports: AI Data Analysis Agent extracts insights from reports
Historical Data: Analyzes trends from historical documents
Research Papers: Summarizes research from academic papers

Knowledge Base

Company Wiki: Searchable company knowledge base
Documentation: Technical documentation Q&A
Best Practices: Answers from best practice documents

Technical Architecture

Craveva AI Enterprise RAG architecture:

Embedding Models: Latest embedding models for semantic understanding
Vector Storage: Optimized vector database for fast retrieval
Search Algorithms: Advanced similarity search algorithms
Caching: Performance optimization through intelligent caching
Chunking Strategy: Optimal chunk sizes for balance of context and retrieval

Benefits

Craveva AI Enterprise RAG provides:

Accuracy: Context-aware responses based on your actual documents
Relevance: Finds right information from large document collections
Speed: Fast retrieval (sub-second) even with thousands of documents
Scalability: Handles large document collections efficiently
No Training Required: Works with your documents immediately, no model training needed

Real-World Example

A restaurant chain uses RAG for staff training:

Setup:

Uploaded 50 SOP documents (PDF) to Google Drive
Connected Google Drive to Craveva AI Enterprise
Platform processed all documents automatically
Created AI Internal Coach agent with RAG enabled

Usage:

Staff ask: "What's the procedure for handling customer complaints?"
RAG retrieves relevant SOP section
AI generates answer based on actual SOP document
Staff get accurate, document-based answers

Results:

92% accuracy on SOP questions
40% faster onboarding (staff get instant answers)
Consistent training across all outlets

Best Practices

Organize Documents: Well-organized documents improve retrieval accuracy
Clear Titles: Descriptive document titles help retrieval
Update Regularly: Keep documents up-to-date for accurate answers
Test Queries: Test common questions to ensure good retrieval
Monitor Performance: Track retrieval accuracy and improve

Conclusion

Craveva AI Enterprise's RAG implementation makes your documents, SOPs, and knowledge bases searchable and accessible to AI agents. By processing documents, creating embeddings, and storing them in vector databases, the platform enables AI agents to answer questions from your actual content with high accuracy. Whether you upload files directly or connect Google Drive, RAG ensures your AI agents have access to your knowledge base, enabling accurate, context-aware responses that improve customer service, staff training, and data analysis.

KPIs to track

Upsell acceptance by menu item and daypart
Ingredient substitution rate and margin impact
Stockout rate, lost sales signals, and substitution frequency
PO approval turnaround and exception rate
Critical incidents: downtime minutes and recovery time
Schedule adherence and overtime variance