Optimize Performance

Step 6: Monitor Performance (future release)

Access your dashboard to track decision accuracy and business impact:

• Decision Accuracy - Track recommendation effectiveness
• User Engagement - Monitor interaction rates
• Conversion Tracking - Measure business impact
• A/B Testing - Compare decision strategies
• Real-time Monitoring - Live performance dashboard

Step 7: Optimize Performance (future release)

You need to optimize performance when

Scenario 1: Proof-of-Concept & A/B Testing

In minutes, you can change or add more data to improve performance. You can also revise instruction to align models with your business goals. Once changed, you can run A/B tests to compare different strategies.

curl -X POST "https://api.trydodo.xyz/api/recommend?project_id=${PROJECT_ID}" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer ${API_KEY}" \
  -d '{
  "template": "Focus on recent purchases to capture recent shopping trend. Please recommend something for me to overcome recent flu. Age: {age}, Gender: {gender}, Location: {location}, Recent purchases: {shopping_history}",
  "context": {
    "age": 28,
    "gender": "male",
    "location": "Dallas",
    "shopping_history": ["winter coat", "thermal shirt", "wool socks", "gloves", "scarf"]
  }
}'

or change instruction for another personalization strategy without modifying data pipelines or model training.

curl -X POST "https://api.trydodo.xyz/api/recommend?project_id=${PROJECT_ID}" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer ${API_KEY}" \
  -d '

{
  "template": "Find something to complement my recent purchases in the same category. My recent purchases: {shopping_history}. Age: {age}, Gender: {gender}, Location: {location}",
  "context": {
    "age": 28,
    "gender": "male",
    "location": "Dallas",
    "shopping_history": ["winter coat", "thermal shirt", "wool socks", "gloves", "scarf"]
  }
}

Scenario 2: Drifting

Drifting is when personalization performance degrades due to shifting preferences of your users over time. You can tackle this by:

• Manual-adjustment: manually add more data or revise instruction as per the scenario 1 instructions. This approach provides more control.
• Auto-adjustment: our AI agent (when enabled) tracks drifting, makes changes, runs A/B tests, and reports to you. This approach provides less control.
• Continue-learning: your model continues learning from collected user signals to adapt to your end users. Such a hands-off approach is best for production.

Optimizing Context and Template

In above code snippets, the template has {context} as a placeholder for the context data. However, this serves as a simple example. In practice, you can design more sophisticated templates to guide the model's recommendations. The more detailed the context and template are, the better the recommendations will be. For example,

• Context:

{
  "user_shopping_history": [
    {
      "product_name": "iPhone 16 512Gb",
      "price": 999,
      "category": "electronics"
    },
    {
      "product_name": "Rich Dad Poor Dad book",
      "price": 10,
      "category": "books"
    }
  ],
  "user_demographic": {
    "age": 25,
    "gender": "male"
  },
  "user_preferences": {
    "preferred_categories": ["electronics", "books"],
    "budget": 100
  }
}

• Template:

{
  "template": "Recommend the next product based on the user's purchase history: {user_shopping_history}, demographics: {user_demographic}, preferences: {user_preferences}. Suggest items under ${user_preferences.budget} that complement these purchases."
}

Modify template and context to fit your specific use case. Check out our industry templates for more examples.