🧪 PharmaOne Pharma Product QnA Assistant

Build an AI-powered assistant for pharmaceutical representatives to answer product questions and check inventory

📋 Overview

This demonstration showcases a complete solution for pharmaceutical representatives who need instant access to accurate product information and real-time inventory data while in the field meeting with healthcare providers.

✅ Key Capabilities

Capability	Description
Product Knowledge	Answer detailed questions about medication indications, contraindications, dosages, and side efects
Inventory Management	Check real-time stock levels across multiple distribution centers
Memory	Maintain conversation context and recall previous interactions
Compliance	Generate accurate responses based only on approved documentation

🏗️ Solution Architecture

Component	Purpose
MAAP Data Loader	Ingest and process pharmaceutical documentation (PDFs, package inserts, technical data)
MAAP Agent Builder	Create a conversational AI interface with specialized tool-calling capabilities
Memory MCP Server	Store and manage conversational memory for contextual responses
MongoDB Atlas	Store and query both unstructured product knowledge and structured inventory data

🔄 Implementation Process

This implementation guide is organized into three sequential phases:

📊 Phase 1: Data Pipeline

Download pharmaceutical product PDFs from PharmaOne
Process documents using MAAP Data Loader
Extract and index meaningful content
Import structured inventory and product data

🤖 Phase 2: Agent Creation

Configure the pharmaceutical representative agent
Set up vector search for product knowledge
Integrate MongoDB tools for inventory queries
Define agent behavior parameters

🧪 Phase 3: Validation

Test with representative medical field queries
Verify accuracy against official documentation
Confirm inventory checking capabilities
Evaluate response quality and compliance

📚 PHASE 1: DATA PIPELINE

Ingest and process PharmaOne pharmaceutical documentation

This phase walks through setting up the data pipeline that will power your pharmaceutical assistant. We'll process product documentation and create searchable collections in MongoDB Atlas.

⚙️ Prerequisites

Before starting, verify you have the following components ready:

Requirement	Details
MongoDB Atlas	Active cluster with connection string
MAAP Data Loader	Installed and configured
Python Environment	Python 3.8+ with required libraries
Source Data	Access to PharmaOne product documentation URLs
AWS Account	Optional: For AWS Bedrock embedding generation

🛠️ Setup Process

Step 1: Environment Configuration

First, set up your development environment:

# Clone the repository
git clone https://github.com/mongodb-partners/maap-data-loader.git
cd maap-data-loader

# Create and activate virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Step 2: Configuration File Setup

Create a .env file in the project root directory:

#===================================
# MongoDB Configuration
#===================================
MONGODB_URI=mongodb+srv://username:password@cluster.mongodb.net
MONGODB_DATABASE=dr_reddys_demo
MONGODB_COLLECTION=product_documents

#===================================
# AWS Configuration
#===================================
AWS_ACCESS_KEY_ID=your_aws_access_key
AWS_SECRET_ACCESS_KEY=your_aws_secret_key
AWS_REGION=us-east-1

#===================================
# Application Settings
#===================================
LOCAL_FILE_OUTPUT_DIR=./processed_files
RUN_ENV=local

Step 3: Document Collection

3.1 Download Product Documentation

Gather PharmaOne pharmaceutical product documentation using the provided scripts:

Python Script (Advanced)

# Navigate to the data directory
cd data

# Install required dependencies
pip install requests pdfkit selenium webdriver-manager

# Execute the Python downloader with options
python download_urls.py --output-dir ./pdf_downloads --concurrent 3

# Return to project root
cd ..

3.2 Document Content Overview

The downloaded PDFs contain comprehensive pharmaceutical information:

Content Type	Description
Active Ingredients	Chemical compounds and their concentrations
Indications	Approved medical uses and conditions
Contraindications	Situations when medication should not be used
Dosage Information	Recommended dosages by age, weight, condition
Side Effects	Common and rare adverse reactions
Storage Instructions	Temperature, humidity and handling requirements
Regulatory Information	Approval details and compliance data

🔄 Data Processing Workflow

Step 4: Document Processing

4.1 Start the Data Processing Service

Launch the MAAP Data Loader service:

# Start the processing server
python app.py

4.2 Verify Service Health

Confirm the service is running correctly:

curl -X GET http://localhost:8000/health

Expected response:

{"status": "healthy"}

4.3 Process Individual Documents

Process a single document to test the pipeline:

# Process a single PDF file
curl -X POST http://localhost:8000/local/upload \
  -H "Content-Type: multipart/form-data" \
  -F "files=@./data/pdf_downloads/all5-17.07.2024.pdf" \
  -F 'json_input_params={
    "user_id": "dr_reddys_demo",
    "mongodb_config": {
      "uri": "mongodb+srv://username:password@cluster.mongodb.net",
      "database": "dr_reddys_demo",
      "collection": "product_documents",
      "index_name": "vector_index",
      "text_field": "text",
      "embedding_field": "embedding"
    }
  }'

4.4 Batch Process All Documents

Create and execute a batch processing script for all documents:

# Create batch processing script
cat > process_all_pdfs.sh << 'EOF'
#!/bin/bash

PDF_DIR="./data/pdf_downloads"
PROCESSED=0
TOTAL=$(find "$PDF_DIR" -name "*.pdf" | wc -l)

echo "🔄 Starting batch processing of $TOTAL PDF files"

for pdf_file in "$PDF_DIR"/*.pdf; do
  if [ -f "$pdf_file" ]; then
    PROCESSED=$((PROCESSED+1))
    FILENAME=$(basename "$pdf_file")
    
    echo "📄 [$PROCESSED/$TOTAL] Processing: $FILENAME"
    
    curl -s -X POST http://localhost:8000/local/upload \
      -H "Content-Type: multipart/form-data" \
      -F "files=@$pdf_file" \
      -F 'json_input_params={
        "user_id": "dr_reddys_demo",
        "mongodb_config": {
          "uri": "mongodb+srv://username:password@cluster.mongodb.net",
          "database": "dr_reddys_demo",
          "collection": "product_documents",
          "index_name": "vector_index",
          "text_field": "text",
          "embedding_field": "embedding"
        }
      }'
      
    echo -e "\n⏱️ Waiting 2 seconds before next file..."
    sleep 2
  fi
done

echo "✅ Completed processing $PROCESSED files"
EOF

# Make the script executable
chmod +x process_all_pdfs.sh

# Run the batch processing script
./process_all_pdfs.sh

4.5 Expected Processing Results

Each successfully processed document will return:

{
  "success": true,
  "message": "Upload processed successfully",
  "details": {
    "files_processed": 1,
    "urls_processed": 0,
    "documents_stored": 1,
    "file_list": ["all5-17.07.2024.pdf"],
    "url_list": []
  }
}

🔍 Verification Process

Step 5: Data Validation

5.1 MongoDB Atlas Web Interface

Verify proper document ingestion using the MongoDB Atlas web interface:

Log in to your MongoDB Atlas account
Navigate to your project > Database > Browse Collections
Select the dr_reddys_demo database
Examine the product_info collection
Confirm documents contain the following structure:

{
  "_id": ObjectId("..."),
  "text": "ALLERWAY 5: COMPOSITION: Each film-coated tablet contains levocetirizine dihydrochloride 5 mg...",
  "embedding": [0.023, -0.128, 0.332, ...],
  "user_id": "dr_reddys_demo",
  "metadata": {
    "filename": "all5-17.07.2024.pdf",
    "processed_at": "2025-08-08T10:23:45Z",
    "pages": 4,
    "file_type": "application/pdf"
  }
}

5.2 MongoDB Compass (Desktop Client)

For more advanced validation using the MongoDB Compass desktop client:

Download and install MongoDB Compass
Connect using your MongoDB Atlas connection string
Navigate to the dr_reddys_demo.product_info collection
Run sample validation queries:

Sample Query 1: Find documents by product name

{text: {$regex: "Cifloc", $options: "i"}}

Sample Query 2: Find documents with dosage information

{text: {$regex: "recommended dose|dosage|administered", $options: "i"}}

Sample Query 3: Find documents with side effects

{text: {$regex: "adverse|side effects|warnings", $options: "i"}}

📊 PHASE 1B: STRUCTURED DATA INTEGRATION

Enhance your agent with real-time inventory and product data

While PDF content provides valuable pharmaceutical information, structured data gives your agent access to precise, queryable business data like inventory levels and product specifications.

🗃️ MongoDB Inventory Integration

Step 1: Prepare Structured Data Schema

Our inventory system will track products across multiple distribution centers with the following information:

Product ID and name
Batch numbers and expiration dates
Quantity available
Distribution center location
Pricing information

Step 2: Run Data Ingestion Script

Execute our structured data ingestion script:

# Navigate to the demo directory
cd dr_reddy_pharma_sales_rep_demo

# Execute the import script
python ingest_structured_data.py \
  --mongodb-uri "mongodb+srv://username:password@cluster.mongodb.net" \
  --database "dr_reddys_demo" \
  --products-collection "products" \
  --inventory-collection "inventory" \
  --centers-collection "distribution_centers"

Step 3: Script Functionality

The ingest_structured_data.py script performs these key operations:

Operation	Description
Data Loading	Imports structured product data from `data/dr_reddys_products.json`
Collection Creation	Creates and populates three interrelated collections
Relationship Building	Establishes connections between products, documents, and inventory
Data Enrichment	Adds computed fields and search optimizations

Step 4: Data Model Overview

The script creates a comprehensive data model with these collections:

Collection: products

{
  "_id": ObjectId("..."),
  "brand_name": "Allerway 5",
  "molecule": "Levocetirizine Dihydrochloride",
  "therapy": "Antihistaminics",
  "dosage_form": "Tabs",
  "package_insert_url": "https://www.drreddys.co.za/sites/.../all5-17.07.2024.pdf",
  "document_id": ObjectId("..."),  // Link to unstructured PDF data
  "search_text": "Allerway 5 Levocetirizine Dihydrochloride"
}

Collection: inventory

{
  "_id": ObjectId("..."),
  "product_id": ObjectId("..."),
  "distribution_center_id": ObjectId("..."),
  "quantity": 315,
  "last_updated": ISODate("2025-08-08T12:34:56Z"),
  "reorder_level": 50,
  "available": true
}

Collection: distribution_centers

{
  "_id": ObjectId("..."),
  "name": "Mumbai Central",
  "location": {
    "city": "Mumbai",
    "state": "Maharashtra",
    "country": "India"
  },
  "address": "123 Main Street, Mumbai",
  "contact": "+91-9876543210"
}

Step 5: Verify Data Ingestion

A successful import will show this output:

🔄 Loading data from ./data/dr_reddys_products.json
✅ Successfully connected to MongoDB Atlas
📊 Created products collection with 38 products
📦 Created inventory collection with 152 inventory records
🏢 Created distribution centers collection with 12 centers
🔗 Successfully linked PDF documents with product data

🔄 Dual-Data Advantage

With both unstructured and structured data sources integrated, your agent can now:

Access Deep Knowledge: Retrieve detailed pharmaceutical information from product documentation
Provide Business Context: Offer real-time inventory levels, pricing, and availability information
Answer Complex Queries: Respond to questions that require both knowledge types:
- "What is the recommended dosage for Allerway 5 and is it available in the Boston distribution center?"
- "Tell me about the side effects of Reegen-D 15000 and check if it's available for expedited shipping from our New York warehouse."

🤖 PHASE 2: AGENT CREATION

Build an intelligent pharmaceutical assistant with MAAP Agent Builder

Now that our data pipeline is complete with both unstructured (PDFs) and structured (products/inventory) data in MongoDB Atlas, we'll install, configure, and deploy a specialized agent for pharmaceutical representatives.

🔧 Installing MAAP Agent Builder

Before configuring your agent, you need to set up the MAAP Agent Builder framework:

# Clone the repository
git clone https://github.com/mongodb-partners/maap-agent-builder.git
cd maap-agent-builder

# Set up a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install the package
pip install -e .

# Create necessary directories
mkdir -p config logs prompts

Required Dependencies

The agent builder requires several key dependencies:

# Install core dependencies for MongoDB integration
pip install pymongo langchain-mongodb

# Install embedding providers
pip install voyageai boto3 langchain-aws

# Install MCP dependencies for memory capabilities
pip install langchain-mcp-adapters fastmcp

🛠️ Agent Configuration

Step 1: YAML Configuration

Create or update your config/agents.yaml file with the following configuration:

#===================================
# PharmaOne Pharma Sales Rep Agent - Configuration
#===================================

# Configure the embedding models
embeddings:
  - name: voyage_embeddings
    provider: voyageai
    model_name: voyage-3.5-lite
    additional_kwargs:
      output_dimension: 512
  - name: bedrock-titan
    provider: bedrock
    model_name: amazon.titan-embed-text-v1

# Configure the language model
llms:
  - name: claude-3.5
    provider: bedrock
    model_name: anthropic.claude-3-5-sonnet-20240620-v1:0
    temperature: 0.7
    max_tokens: 4096

# Configure agent tools
tools:
  # Product Availability Tool (Full-text search)
  - name: product_availability
    tool_type: full_text_search
    description: Checks product availability of the product in the MongoDB database
    connection_str: ${MONGODB_URI:-mongodb://localhost:27017}
    namespace: dr_reddys.product_inventory
    additional_kwargs:
      index_name: full_text_search_index
      text_field: search_text
      top_k: 5
      
  # Product Information Tool (Vector search)
  - name: product_information
    tool_type: vector_search
    description: Searches for relevant documents in the vector store to help with FAQ on pharmaone product information
    connection_str: ${MONGODB_URI:-mongodb://localhost:27017}
    namespace: dr_reddys.product_info
    embedding_model: voyage_embeddings  # Reference to the embedding model defined above
    additional_kwargs:
      index_name: vector_index
      top_k: 3
      min_score: 0.7
      
  # AI Memory Tool (MCP server integration)
  - name: ai-memory
    tool_type: mcp
    description: MongoDB MCP Toolkit for advanced data operations
    servers_config:
      "ai-memory":
        url: "http://localhost:8080/mcp"
        transport: "streamable_http"
        filter:
          - store_memory
          - retrieve_memory
          - semantic_cache_response

# Configure checkpointing
checkpointer:
  connection_str: ${MONGODB_URI:-mongodb://localhost:27017}
  db_name: dr_reddy_agent_state
  collection_name: checkpoints
  name: rag_agent_checkpointer

# Configure the agent
agent:
  name: rag_react_agent
  agent_type: react
  llm: claude-3.5  # Reference to the LLM defined above
  tools:
    - product_information  # References to the tools defined above
    - product_availability
    - ai-memory
  system_prompt_path: ./prompts/rag_system_prompt.txt

Step 2: Agent Deployment

Launch the agent service with all required environment variables:

# Set required environment variables
export MONGODB_URI="mongodb+srv://username:password@cluster.mongodb.net"
export AWS_ACCESS_KEY_ID="your_aws_access_key"
export AWS_SECRET_ACCESS_KEY="your_aws_secret_key"
export AWS_REGION="us-east-1"
export VOYAGE_API_KEY="your_voyage_api_key"  # Required for voyage embeddings

# Start the agent service
python -m agent_builder.cli serve \
  --config config/agents.yaml \
  --host 0.0.0.0 \
  --port 5000

Step 3: Create System Prompt

Create a system prompt file that defines your agent's behavior and capabilities:

# Create the prompts directory if it doesn't exist
mkdir -p prompts

# Create the system prompt file
cat > prompts/rag_system_prompt.txt << 'EOF'
# Pharmaceutical Sales Representative Assistant

You are an AI assistant specialized in helping pharmaceutical sales representatives with detailed product information, inventory management, and clinical knowledge.

## Your Capabilities

- Access detailed pharmaceutical product information including indications, contraindications, dosages, side effects
- Check real-time inventory levels across distribution centers
- Maintain conversation memory to provide contextual assistance
- Generate compliant, factual responses based only on approved documentation

## Guidelines

1. Always provide factually accurate information based on official PharmaOne product documentation
2. When discussing medications, include important safety information and contraindications
3. For inventory queries, provide specific details on availability, location, and quantity
4. Use professional, clear language appropriate for pharmaceutical professionals
5. Maintain conversation context by referencing relevant details from previous exchanges
6. If unsure about any information, clearly state the limitations of your knowledge
7. Do not provide medical advice beyond what is contained in official product documentation

Remember: Your responses will be used by pharmaceutical representatives when speaking with healthcare professionals, so accuracy and compliance are essential.
EOF

Step 4: Agent Validation

Test the agent through the API:

# Example API request
curl -X POST http://localhost:5000/chat \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {"role": "user", "content": "What are the indications for Allerway 5?"}
    ],
    "config": {
      "thread_id": "user-session-123",
      "recursion_limit": 10,
    }
  }'

Or access the web interface by opening: http://localhost:5000

📝 Example Prompts for Testing

Test your agent with these pharmaceutical representative scenarios:

Category	Example Prompt
Product Information	"What are the side effects of Cifloc and how do they compare to similar antibiotics?"
Dosage Guidance	"A doctor is considering prescribing Alnedra to an elderly patient with kidney issues. What's the recommended dosage adjustment?"
Inventory Check	"I'm meeting with a hospital pharmacy manager tomorrow. What's our current inventory level of Drymed 500 in the Mumbai distribution center?"
Safety Information	"A patient is pregnant in her second trimester. Are there any contraindications for using Cifloc in her case?"
Storage Requirements	"A clinic in a rural area without reliable refrigeration asks about Lenap 10mg. What are the storage requirements?"
Combined Query	"I need information about Reegen-D 15000 for a dermatology clinic. What are the key benefits and do we have enough stock to fulfill a large order?"

� PHASE 2B: MCP SERVER INTEGRATION

Enhance your agent with MongoDB Memory MCP server integration

After configuring the basic agent, you can extend its capabilities by integrating with the MongoDB Memory MCP server, providing advanced memory management, semantic caching, and hybrid search functionality.

🛠️ Memory MCP Setup

The MongoDB Memory MCP server provides powerful memory management for conversational agents, allowing them to store and retrieve information across sessions.

Step 1: Clone the Memory MCP Repository

# Clone the MongoDB Memory MCP repository
git clone https://github.com/mongodb-partners/memory-mcp.git
cd memory-mcp

# Create necessary environment files from examples
cp .env.example .env

Step 2: Configure Environment Variables

Edit the .env file with your MongoDB and AWS credentials:

# Core Application Settings
APP_NAME=memory-mcp
DEBUG=False
PORT=8080

# Service URLs (default Docker Compose settings)
LOGGER_SERVICE_URL=http://event-logger:8181
AI_MEMORY_SERVICE_URL=http://ai-memory:8182
SEMANTIC_CACHE_SERVICE_URL=http://semantic-cache:8183

# MongoDB Configuration
MONGODB_URI=mongodb+srv://username:password@cluster.mongodb.net
DATABASE_NAME=memory_mcp
COLLECTION_NAME=memory

# AWS Configuration
AWS_ACCESS_KEY_ID=your_access_key
AWS_SECRET_ACCESS_KEY=your_secret_key
AWS_REGION=us-east-1
EMBEDDING_MODEL_ID=amazon.titan-embed-text-v1
LLM_MODEL_ID=us.anthropic.claude-3-sonnet-20240229-v1:0
VECTOR_DIMENSION=1536

# External APIs
TAVILY_API_KEY=your_tavily_api_key

Step 3: Deploy Using Docker Compose

# Build and start all services
docker-compose up -d

# Verify deployment
docker-compose ps
docker-compose logs memory-mcp

💡 Integrating Memory MCP with Agent Builder

Step 1: Update Agent Configuration

Your config/agents.yaml file already includes the Memory MCP integration with the ai-memory tool. Here's the relevant section highlighting the MCP configuration:

# AI Memory Tool Configuration
tools:
  # Other tools configuration...
  
  # Memory MCP Integration
  - name: ai-memory
    tool_type: mcp
    description: MongoDB MCP Toolkit for advanced data operations
    servers_config:
      "ai-memory":
        url: "http://localhost:8080/mcp"
        transport: "streamable_http"
        filter:
          - store_memory
          - retrieve_memory
          - semantic_cache_response

# Agent configuration using the Memory MCP tool
agent:
  name: rag_react_agent
  agent_type: react
  llm: claude-3.5
  tools:
    - product_information
    - product_availability
    - ai-memory  # AI Memory tool is included in the agent's toolset
  system_prompt_path: ./prompts/rag_system_prompt.txt

Step 2: Understanding Memory MCP Tools

The Memory MCP server provides several powerful tools that enhance your pharmaceutical sales rep agent:

Tool	Purpose	Benefits
store_memory	Save conversation messages for future context	Maintains conversation history across sessions
retrieve_memory	Get relevant context from past conversations	Provides continuity for pharmaceutical discussions
semantic_cache_response	Cache responses for similar queries	Reduces latency for common drug inquiries
check_semantic_cache	Check for existing similar responses	Ensures consistency in medication information
hybrid_search	Advanced document search combining vector and keyword matching	Improves accuracy of pharmaceutical documentation searches

Step 3: Testing Memory-Enhanced Agent

Start your agent with the updated configuration:

# Start the agent builder service with MCP integration
python -m agent_builder.cli serve \
  --config config/agents.yaml \
  --host 0.0.0.0 \
  --port 5000

Test the memory capabilities with sequential queries:

First query establishes context:

"Tell me about the side effects of Allerway 5"

Follow-up query references previous context:

"Given those side effects, what precautions should be taken?"

The agent should respond using memory from the previous interaction, demonstrating continuity of conversation.

�🚀 PHASE 3: VALIDATION & EXPANSION

Test, validate and expand your pharmaceutical assistant

After completing the configuration and deployment of your pharmaceutical sales representative agent, it's essential to thoroughly test its capabilities and plan for future enhancements.

🧪 Agent Testing Strategy

Step 1: Knowledge Testing

Verify the agent can accurately retrieve information from pharmaceutical documentation:

Factual Queries
- "What is the active ingredient in Allerway 5?"
- "What are the contraindications for Cifloc?"
- "What is the recommended pediatric dosage for Lenap 10mg?"
Contextual Understanding
- "Compare the side effects of Drymed 500 and Cifloc"
- "Explain when Alnedra should be taken with food versus without"
- "What safety precautions should be taken with Reegen-D 15000?"

Step 2: Inventory Intelligence

Test the agent's ability to access structured inventory data:

Basic Inventory Queries
- "How many units of Allerway 5 are available across all distribution centers?"
- "Is Cifloc currently in stock at the Mumbai distribution center?"
- "What's our current inventory level for Drymed 500?"
Complex Business Logic
- "Which distribution center has the highest stock of Lenap 10mg?"
- "Are we below the reorder threshold for Alnedra in any location?"
- "What's the total inventory value of antihistamine products?"

🔄 Continuous Improvement

Enhancement Roadmap

Enhancement	Description	Priority
Image Recognition	Allow scanning of product packaging	Medium
Prescription Analysis	Extract information from prescription images	High
Mobile Integration	Native mobile app for field representatives	High
Offline Mode	Limited functionality without internet connection	Medium
Competitor Analysis	Add comparative product information	Low

Deployment Options

Cloud Deployment
- Deploy to MongoDB Atlas App Services
- Configure with AWS/Azure/GCP serverless functions
- Set up CI/CD pipeline for automated updates
On-Premises Solutions
- Kubernetes deployment within corporate infrastructure
- Integration with existing enterprise systems
- Compliance with internal security policies
Hybrid Approaches
- Keep sensitive data on-premises
- Use cloud resources for model serving
- Implement proper data encryption and access controls

🔍 Troubleshooting Guide

If you encounter issues during setup or operation, here are some common problems and solutions:

MongoDB Connection Issues

Error: MongoDB connection failed

Solutions:

Verify your MongoDB Atlas connection string
Check if your IP address is whitelisted in MongoDB Atlas
Ensure your cluster is running and accessible
Test connection with MongoDB Compass

Embedding Model Errors

Error: Failed to generate embeddings

Solutions:

Verify your API keys for VoyageAI or AWS are correct and active
Check that the embedding model specified exists
Ensure your network can reach the embedding service
Try reducing embedding dimensions if you encounter memory issues

MCP Server Connection Problems

Error: Failed to connect to MCP server

Solutions:

Verify the MCP server is running with docker-compose ps
Check if the port is correctly exposed and accessible
Review the MCP server logs with docker-compose logs memory-mcp
Ensure your MongoDB configuration for the MCP server is correct

Agent Execution Errors

Error: Tool execution failed

Solutions:

Check if all required environment variables are set
Verify namespace values match your MongoDB collections
Review tool permissions and settings
Test each tool individually to isolate the problem

📋 Overview​

✅ Key Capabilities​

🏗️ Solution Architecture​

🔄 Implementation Process​

📊 Phase 1: Data Pipeline​

🤖 Phase 2: Agent Creation​

🧪 Phase 3: Validation​

📚 PHASE 1: DATA PIPELINE

⚙️ Prerequisites​

🛠️ Setup Process​

Step 1: Environment Configuration​

Step 2: Configuration File Setup​

Step 3: Document Collection​

3.1 Download Product Documentation​

3.2 Document Content Overview​

🔄 Data Processing Workflow​

Step 4: Document Processing​

4.1 Start the Data Processing Service​

4.2 Verify Service Health​

4.3 Process Individual Documents​

4.4 Batch Process All Documents​

4.5 Expected Processing Results​

🔍 Verification Process​

Step 5: Data Validation​

5.1 MongoDB Atlas Web Interface​

5.2 MongoDB Compass (Desktop Client)​

📊 PHASE 1B: STRUCTURED DATA INTEGRATION

🗃️ MongoDB Inventory Integration​

Step 1: Prepare Structured Data Schema​

Step 2: Run Data Ingestion Script​

Step 3: Script Functionality​

Step 4: Data Model Overview​

Step 5: Verify Data Ingestion​

🔄 Dual-Data Advantage​

🤖 PHASE 2: AGENT CREATION

🔧 Installing MAAP Agent Builder​

Required Dependencies​

🛠️ Agent Configuration​

Step 1: YAML Configuration​

Step 2: Agent Deployment​

Step 3: Create System Prompt​

Step 4: Agent Validation​

📝 Example Prompts for Testing​

� PHASE 2B: MCP SERVER INTEGRATION

🛠️ Memory MCP Setup​

Step 1: Clone the Memory MCP Repository​

Step 2: Configure Environment Variables​

Step 3: Deploy Using Docker Compose​

💡 Integrating Memory MCP with Agent Builder​

Step 1: Update Agent Configuration​

Step 2: Understanding Memory MCP Tools​

Step 3: Testing Memory-Enhanced Agent​

�🚀 PHASE 3: VALIDATION & EXPANSION

🧪 Agent Testing Strategy​

Step 1: Knowledge Testing​

Step 2: Inventory Intelligence​

🔄 Continuous Improvement​

Enhancement Roadmap​

Deployment Options​

🔍 Troubleshooting Guide​

MongoDB Connection Issues​

Embedding Model Errors​

MCP Server Connection Problems​

Agent Execution Errors​

📋 Overview

✅ Key Capabilities

🏗️ Solution Architecture

🔄 Implementation Process

📊 Phase 1: Data Pipeline

🤖 Phase 2: Agent Creation

🧪 Phase 3: Validation

⚙️ Prerequisites

🛠️ Setup Process

Step 1: Environment Configuration

Step 2: Configuration File Setup

Step 3: Document Collection

3.1 Download Product Documentation

3.2 Document Content Overview

🔄 Data Processing Workflow

Step 4: Document Processing

4.1 Start the Data Processing Service

4.2 Verify Service Health

4.3 Process Individual Documents

4.4 Batch Process All Documents

4.5 Expected Processing Results

🔍 Verification Process

Step 5: Data Validation

5.1 MongoDB Atlas Web Interface

5.2 MongoDB Compass (Desktop Client)

🗃️ MongoDB Inventory Integration

Step 1: Prepare Structured Data Schema

Step 2: Run Data Ingestion Script

Step 3: Script Functionality

Step 4: Data Model Overview

Step 5: Verify Data Ingestion

🔄 Dual-Data Advantage

🔧 Installing MAAP Agent Builder

Required Dependencies

🛠️ Agent Configuration

Step 1: YAML Configuration

Step 2: Agent Deployment

Step 3: Create System Prompt

Step 4: Agent Validation

📝 Example Prompts for Testing

🛠️ Memory MCP Setup

Step 1: Clone the Memory MCP Repository

Step 2: Configure Environment Variables

Step 3: Deploy Using Docker Compose

💡 Integrating Memory MCP with Agent Builder

Step 1: Update Agent Configuration

Step 2: Understanding Memory MCP Tools

Step 3: Testing Memory-Enhanced Agent

🧪 Agent Testing Strategy

Step 1: Knowledge Testing

Step 2: Inventory Intelligence

🔄 Continuous Improvement

Enhancement Roadmap

Deployment Options

🔍 Troubleshooting Guide

MongoDB Connection Issues

Embedding Model Errors

MCP Server Connection Problems

Agent Execution Errors