Assistant Profiles

Overview

Assistant Profiles let you define how embedded AI assistants behave within the context of each dashboard. They control how the assistant understands user questions, queries and analyzes data, and communicates insights, ensuring every response aligns with your specific business logic and data structure.

By attaching a profile to a dashboard, you create a purpose-built assistant that delivers accurate, relevant, and context-aware answers. Profiles are fully customizable, reflect your business terminology and metric definitions, and can be reused or adapted across multiple dashboards to support different teams or use cases.

---

What Is an Assistant Profile?

An Assistant Profile includes:

• Custom Instructions Logic and definitions that guide the assistant in understanding your business vocabulary, applying consistent metric definitions, and structuring SQL queries and joins.

• Seed Questions
  Seed Questions are examples that appear as placeholders when a user opens the assistant on a dashboard. They act as conversation starters, helping end users understand the types of questions they can ask and how to begin interacting with the assistant.

• Model Settings
  Parameters such as temperature (creativity) and token limits (response length)

Once configured, you can attach the profile to a dashboard to customize how the assistant behaves in that context.

---

Custom Instructions for Business Vocabulary

Custom instructions are essential to every Assistant Profile. They let you define how the assistant reasons and communicates based on your domain logic:

Defining Business Concepts

Describe terms like active customer, churn, or high-value account.
E.g.: “An active user is one who logged in at least 3 times in the last 30 days.”

Explaining Metric Logic

Express how KPIs like Gross Margin or Net Revenue Retention are calculated, including joins and filters.
E.g.: “Net Revenue Retention = (Current MRR from existing customers) ÷ (MRR from those customers 12 months ago).”

Clarifying Join Relationships

Explain how to join tables to build composite metrics.
E.g.: “To calculate average order value, join orders and order_items on order_id and divide total revenue by total orders.”
E.g.: “To calculate CLTV, join orders with customers on customer_id, group by customer, and sum order values.”

These instructions help the assistant translate natural language into reliable SQL and Python-based analysis—keeping responses consistent with your data model.

---

System Behavior (Core Instructions)

All Assistant Profiles are grounded in structured behaviors that ensure reliability:

• Analytical Thinking: Step-by-step reasoning before answering
• Data Handling: Uses SQL with row limits and Python (e.g., pandas, statsmodels) for deeper analysis
• Chart Support: Visualizes insights using built-in charting tools
• Dynamic Formatting: Regenerates charts and tables to respect user formatting requests

Example:
“Show me revenue trends vs forecast” → Assistant writes SQL, analyzes in Python, and returns a chart + narrative summary.

---

Workflow: Attaching a Profile to a Dashboard

1. Create an Assistant Profile
   Define seed questions, model behavior, and required custom instructions.

2. Attach to a Dashboard
   Link the profile directly to your dashboard—this determines how the assistant responds in that context.

3. Ask Natural Language Questions
   Users can ask about trends, KPIs, or drivers using everyday language—no SQL needed.

4. Receive Insightful Answers
   The assistant queries your data, runs calculations, and returns visual summaries—guided by your definitions.

---

Example Use Cases

---

Configuration Steps

1. Create a Profile

   • Navigate to the Project page → go to the Assistants tab → click Create Profile
   • Define the assistant's name, custom instructions, seed questions, and model settings
   • Save the profile

2. Link to a Dashboard

   • Open the dashboard
   • Expand the assistant
   • Select Change Default Profile and attach the profile you created

3. Start Exploring

   • The assistant will now respond based on the new profile
   • Ask questions using natural language and receive tailored, insightful answers

---