Day 19 challenge

Goal: build document-based retrieval with MongoDB for unstructured data retrievalTheme: context engineering week - NoSQL and document retrievalTime investment: ~25 minutes
Welcome to Day 19! Yesterday you built structured RAG systems with PostgreSQL. Today you’ll work with unstructured document retrieval using MongoDB Atlas. You’ll learn to search through product reviews, feedback, and content where the structure varies and context is embedded in natural language. Document-based retrieval excels when dealing with varied, unstructured content that doesn’t fit neatly into database tables.

​
What you’ll accomplish today

  • Set up MongoDB Atlas with vector search capabilities for document retrieval
  • Load Amazon product reviews as example unstructured data
  • Connect your agent to MongoDB for intelligent document search
  • Build retrieval patterns for sentiment analysis and contextual understanding
This requires setting up MongoDB Atlas (free tier available). You’ll be working with unstructured documents that require different indexing and search strategies than structured data.

​
Step 1: Understanding document-based retrieval

Document databases handle unstructured data differently than relational databases:

​
Document and relational data

Relational data (Day 18):
  • Fixed schema with defined columns
  • Structured relationships between tables
  • Efficient for transactional operations
  • Great for precise, attribute-based queries
Document data (Today):
  • Flexible schema with nested objects and arrays
  • Self-contained documents with varying structures
  • Efficient for content storage and retrieval
  • Great for text analysis and semantic search

​
When to use document-based retrieval

Ideal for:
  • Product reviews and customer feedback
  • Content analysis (blogs, articles, documentation)
  • Social media posts and comments
  • Support tickets and conversation logs
  • Research papers and knowledge articles
Not ideal for:
  • Highly structured transactional data
  • Complex multi-table relationships
  • Frequent schema changes requiring migrations
Document thinking Think of document retrieval as searching through a library of books rather than looking up entries in a catalog. Context and content matter more than exact structure.
MongoDB Atlas provides managed MongoDB with built-in vector search:

​
Create your MongoDB Atlas cluster

  1. Visit mongodb.com/atlas and create a free account
  2. Create a new cluster (M0 free tier is sufficient for learning)
  3. Set up database access with username/password authentication
  4. Configure network access to allow connections from your IP
  5. Get your connection string for later use

​
Create the reviews collection

// Connect to your MongoDB instance and create the reviews collection
use product_reviews

// Create collection with validation schema
db.createCollection("reviews", {
  validator: {
    $jsonSchema: {
      bsonType: "object",
      required: ["product_id", "product_title", "review_text", "rating"],
      properties: {
        product_id: { bsonType: "string" },
        product_title: { bsonType: "string" },
        review_text: { bsonType: "string" },
        rating: { bsonType: "number", minimum: 1, maximum: 5 },
        reviewer_name: { bsonType: "string" },
        review_date: { bsonType: "date" },
        verified_purchase: { bsonType: "bool" },
        helpful_votes: { bsonType: "number" },
        embedding: { bsonType: "array" }
      }
    }
  }
})

​
Load Amazon product reviews data

Create a realistic dataset of product reviews with varying structures:

​
Sample review documents

// Insert sample Amazon-style product reviews
db.reviews.insertMany([
  {
    product_id: "iphone15pro",
    product_title: "Apple iPhone 15 Pro",
    review_text:
      "Absolutely love this phone! The camera quality is incredible, especially in low light. The titanium build feels premium and the battery easily lasts all day. The Action Button is a nice touch. Only downside is the price, but you get what you pay for with Apple.",
    rating: 5,
    reviewer_name: "TechEnthusiast2024",
    review_date: new Date("2024-10-15"),
    verified_purchase: true,
    helpful_votes: 23,
    features_mentioned: ["camera", "battery", "titanium", "Action Button"],
    sentiment: "positive",
  },
  {
    product_id: "iphone15pro",
    product_title: "Apple iPhone 15 Pro",
    review_text:
      "Camera is good but not revolutionary. Battery life is decent but nothing special. The phone feels nice but is way overpriced for what you get. Android flagships offer better value. Not impressed with iOS limitations either.",
    rating: 3,
    reviewer_name: "AndroidUser42",
    review_date: new Date("2024-11-02"),
    verified_purchase: true,
    helpful_votes: 8,
    features_mentioned: ["camera", "battery", "price", "iOS"],
    sentiment: "neutral",
  },
  {
    product_id: "galaxy_s24_ultra",
    product_title: "Samsung Galaxy S24 Ultra",
    review_text:
      "The S Pen integration is fantastic for note-taking and creative work. Display is gorgeous and the camera zoom capabilities are unmatched. AI features are actually useful, especially for photo editing. Great phone for productivity.",
    rating: 5,
    reviewer_name: "ProductivityPro",
    review_date: new Date("2024-09-28"),
    verified_purchase: true,
    helpful_votes: 15,
    features_mentioned: ["S Pen", "display", "camera zoom", "AI features"],
    sentiment: "positive",
  },
  {
    product_id: "sony_wh1000xm5",
    product_title: "Sony WH-1000XM5 Headphones",
    review_text:
      "These headphones are a game changer for my daily commute. Noise cancellation is outstanding - completely blocks out subway noise. Sound quality is crisp and balanced. Comfort is excellent for long wearing sessions. Battery life is as advertised. Highly recommend!",
    rating: 5,
    reviewer_name: "CommuterLife",
    review_date: new Date("2024-10-20"),
    verified_purchase: true,
    helpful_votes: 31,
    features_mentioned: [
      "noise cancellation",
      "sound quality",
      "comfort",
      "battery life",
    ],
    sentiment: "positive",
  },
  {
    product_id: "dyson_v15",
    product_title: "Dyson V15 Detect Vacuum",
    review_text:
      "The laser dust detection is amazing - you can see exactly where dust is hiding. Suction power is incredible on all surfaces. However, it's quite heavy for extended use and the battery drains faster than expected on max power. Worth it but has limitations.",
    rating: 4,
    reviewer_name: "CleaningExpert",
    review_date: new Date("2024-11-05"),
    verified_purchase: true,
    helpful_votes: 12,
    features_mentioned: [
      "laser detection",
      "suction power",
      "weight",
      "battery",
    ],
    sentiment: "mostly_positive",
  },
  {
    product_id: "nike_air_max_270",
    product_title: "Nike Air Max 270",
    review_text:
      "Comfortable for casual wear but not great for serious running. The cushioning feels good for walking around the city. Style is nice and goes with most outfits. Sizing runs a bit large - order half a size down. Good for the price point.",
    rating: 4,
    reviewer_name: "CasualRunner",
    review_date: new Date("2024-10-08"),
    verified_purchase: true,
    helpful_votes: 7,
    features_mentioned: ["comfort", "cushioning", "style", "sizing"],
    sentiment: "positive",
  },
  {
    product_id: "instant_pot_duo",
    product_title: "Instant Pot Duo 7-in-1",
    review_text:
      "This has revolutionized my meal prep! Makes perfect rice, tender meats, and great yogurt. Learning curve was steep initially but now I use it almost daily. Saves so much time compared to traditional cooking methods. Essential kitchen appliance.",
    rating: 5,
    reviewer_name: "MealPrepMaster",
    review_date: new Date("2024-09-15"),
    verified_purchase: true,
    helpful_votes: 45,
    features_mentioned: ["meal prep", "rice", "yogurt", "time saving"],
    sentiment: "positive",
  },
  {
    product_id: "macbook_pro_m3",
    product_title: "MacBook Pro 16-inch M3",
    review_text:
      "Performance is absolutely incredible for video editing and development work. The M3 chip handles everything I throw at it without breaking a sweat. Display is stunning. Battery life is impressive for such a powerful machine. Premium build quality as expected from Apple.",
    rating: 5,
    reviewer_name: "VideoEditor_Pro",
    review_date: new Date("2024-11-10"),
    verified_purchase: true,
    helpful_votes: 28,
    features_mentioned: [
      "performance",
      "M3 chip",
      "video editing",
      "display",
      "battery",
    ],
    sentiment: "positive",
  },
])

​
Connect MongoDB to your agent

Integrate MongoDB with your Hypermode agent for document retrieval:

​
Add MongoDB connection

  1. Navigate to your agent’s connections in the About section
  2. Add MongoDB connection with your Atlas cluster credentials
  3. Test the connection by querying the reviews collection

​
Create a review analysis agent

If needed, create a specialized agent with Concierge: 
I want to create a product review analysis agent that helps users understand customer sentiment and experiences.

The agent should:
- Search through product reviews using semantic similarity
- Analyze customer sentiment and common themes
- Identify specific product strengths and weaknesses mentioned by reviewers
- Compare customer experiences across similar products
- Provide insights about product satisfaction patterns
- Help users understand real customer experiences beyond marketing claims

The agent should act like a market research analyst who specializes in customer feedback analysis.

​
Configure document retrieval patterns

Add these instructions to your agent: 
Review Analysis Guidelines:
1. Use semantic search to find relevant reviews based on user interests
2. Look for patterns in customer feedback, not just individual opinions
3. Identify specific product features mentioned by multiple reviewers
4. Consider both positive and negative feedback for balanced insights
5. Explain sentiment trends and common themes across reviews
6. Provide specific quotes from reviews to support analysis
7. Note verified purchases vs. unverified reviews when relevant

​
Implement document search and analysis

Test your document-based retrieval system: 
What do customers say about camera quality in smartphone reviews? I'm particularly interested in low-light performance.
Your agent should:
  1. Search reviews for camera-related content
  2. Extract specific mentions of camera features and performance
  3. Analyze sentiment around camera quality
  4. Provide quotes from actual reviews
  5. Summarize trends across multiple customer experiences

​
Sentiment analysis queries

Analyze customer satisfaction patterns for headphones. What are the most common complaints and praise points?
This tests the agent’s ability to:
  • Aggregate sentiment across multiple reviews
  • Identify common themes in feedback
  • Distinguish between different types of issues or praise
  • Provide actionable insights from unstructured feedback

​
Comparative analysis

Compare customer experiences between iPhone and Samsung phones based on actual user reviews. What are the key differences in satisfaction?
This requires:
  • Cross-product analysis using semantic search
  • Pattern recognition across different product categories
  • Balanced reporting of pros and cons
  • Context-aware insights about user preferences
Document retrieval insight The power of document-based RAG is in understanding context and nuance that structured data misses. Focus on semantic meaning over exact keyword matches.

​
What you’ve accomplished

In 25 minutes, you’ve mastered document-based RAG systems: MongoDB foundation: set up Atlas with vector search for document retrieval Unstructured data handling: loaded and indexed product reviews with flexible schemas Document search: implemented search across varying content structures Sentiment analysis: built patterns for understanding customer feedback and satisfaction trends Advanced aggregation: explored complex queries for multi-dimensional document analysis

​
The power of document-based retrieval

Document RAG unlocks insights from unstructured content: Before document RAG Limited to structured product information and marketing claims After document RAG Access to real customer experiences, sentiment trends, and nuanced feedback Combined with yesterday’s structured RAG, you now have comprehensive information retrieval capabilities.

Tomorrow - Day 20

Explore GraphRAG with Neo4j for complex relationship reasoning and knowledge discovery through connected data.

​
Pro tip for today

Test document retrieval with nuanced queries: 
Test my document RAG system with these complex queries:
1. "What specific issues do customers mention about battery life across different devices?"
2. "How do customer experiences vary between verified and unverified purchasers?"
3. "What patterns emerge when customers compare competing products?"
4. "Which product features generate the most emotional responses in reviews?"

How well does the system understand context and extract insights from unstructured text?
This reveals the sophistication of your document-based retrieval system.
Time to complete: ~25 minutes Skills learned MongoDB Atlas setup, document-based retrieval, semantic search across unstructured data, sentiment analysis, advanced aggregation patterns Next: day 20 - GraphRAG with Neo4j for relationship-based knowledge discovery
Remember: document-based RAG excels at capturing human context, emotion, and nuanced experiences that structured data often misses. Use it for understanding “why” not just “what.”