AI-Powered Analysis
The most powerful data science workflows combine traditional analytics with AI. Flow-Like lets you build AI agents that can query databases, analyze data, and generate insights—all through natural language.
Why AI + Data Science?
Section titled “Why AI + Data Science?”| Traditional Approach | AI-Powered Approach |
|---|---|
| Write SQL queries manually | Ask questions in plain English |
| Build fixed dashboards | Generate dynamic insights |
| Code data transformations | Describe what you need |
| Static reports | Conversational exploration |
The Data Science Agent
Section titled “The Data Science Agent”A data science agent has access to your data and can:
- Query databases using SQL
- Analyze results and identify patterns
- Create visualizations as charts
- Train ML models and make predictions
- Explain findings in plain language
Building a Data Analysis Agent
Section titled “Building a Data Analysis Agent”Step 1: Create the Agent
Section titled “Step 1: Create the Agent”Make Agent │ ├── Model: (a capable model like GPT-4 or Claude) │ └── Agent ──▶ (agent object)Step 2: Set System Prompt
Section titled “Step 2: Set System Prompt”Set Agent System Prompt │ ├── Agent: (from step 1) ├── System Prompt: │ "You are a data analyst assistant. You have access to: │ - SQL tools to query the data warehouse │ - Visualization tools to create charts │ │ When analyzing data: │ 1. First understand what tables are available │ 2. Write SQL to answer the user's question │ 3. Summarize findings in plain language │ 4. Create visualizations when helpful │ │ Always explain your reasoning." │ └── Agent ──▶ (configured agent)Step 3: Add SQL Tools
Section titled “Step 3: Add SQL Tools”Give the agent access to your DataFusion session:
Add SQL Session │ ├── Agent: (configured agent) ├── Session: (DataFusion session with tables) │ └── Agent ──▶ (agent with SQL access)Step 4: Add Custom Tools (Optional)
Section titled “Step 4: Add Custom Tools (Optional)”Add your own Flow-Like flows as tools:
Add Flow Tools │ ├── Agent: (agent) ├── Flows: [ │ "create_chart", │ "train_model", │ "export_report" │ ] │ └── Agent ──▶ (agent with custom tools)Step 5: Run the Agent
Section titled “Step 5: Run the Agent”Invoke Agent Streaming │ ├── Agent: (fully configured) ├── History: "What were our top 5 products last quarter?" │ ├── On Chunk ──▶ Push Chunk (stream response) │ └── Done ──▶ Final responseExample Conversation
Section titled “Example Conversation”User: “What were our sales trends last month?”
Agent thinks: I need to query the sales table, aggregate by day, and analyze the trend.
Agent executes:
SELECT DATE_TRUNC('day', order_date) as day, SUM(amount) as daily_salesFROM ordersWHERE order_date >= '2025-01-01' AND order_date < '2025-02-01'GROUP BY 1ORDER BY 1Agent responds:
“Sales in January showed an upward trend, starting at $45,000 on January 1st and reaching $78,000 by January 31st—a 73% increase. The strongest days were Fridays, averaging $65,000 compared to $42,000 on Mondays.
Here’s the daily trend:
SQL Tools in Detail
Section titled “SQL Tools in Detail”When you add a SQL session to an agent, it gains these capabilities:
Execute SQL
Section titled “Execute SQL”The agent can run any SQL query:
SELECT customer_segment, COUNT(*) as customers, AVG(lifetime_value) as avg_ltvFROM customersGROUP BY customer_segmentORDER BY avg_ltv DESCExplore Schema
Section titled “Explore Schema”The agent can discover what tables and columns exist:
-- What tables are available?SHOW TABLES
-- What columns are in this table?DESCRIBE salesComplex Analytics
Section titled “Complex Analytics”Window functions, CTEs, joins—the agent can write sophisticated queries:
WITH monthly_sales AS ( SELECT DATE_TRUNC('month', date) as month, product_category, SUM(revenue) as revenue FROM sales GROUP BY 1, 2)SELECT month, product_category, revenue, revenue - LAG(revenue) OVER ( PARTITION BY product_category ORDER BY month ) as month_over_month_changeFROM monthly_salesCreating Tool Flows for Agents
Section titled “Creating Tool Flows for Agents”Build custom capabilities as Flow-Like flows:
Chart Generation Tool
Section titled “Chart Generation Tool”┌────────────────────────────────────────────────────────────┐│ Flow: create_chart ││ ││ Inputs: ││ - data (string): CSV data ││ - chart_type (string): bar, line, pie, etc. ││ - title (string): Chart title ││ ││ Flow: ││ Format Markdown ──▶ Return chart block ││ ││ Output: ││ - chart (string): Markdown with nivo/plotly block ││ │└────────────────────────────────────────────────────────────┘ML Prediction Tool
Section titled “ML Prediction Tool”┌────────────────────────────────────────────────────────────┐│ Flow: predict_churn ││ ││ Inputs: ││ - customer_id (string): Customer to predict for ││ ││ Flow: ││ Lookup Customer ──▶ Load Model ──▶ Predict ││ ││ Output: ││ - prediction (object): {churn_risk: 0.75, factors: []} ││ │└────────────────────────────────────────────────────────────┘Report Export Tool
Section titled “Report Export Tool”┌────────────────────────────────────────────────────────────┐│ Flow: export_report ││ ││ Inputs: ││ - title (string): Report title ││ - content (string): Report markdown ││ - format (string): pdf, csv, html ││ ││ Flow: ││ Generate Report ──▶ Save to Storage ──▶ Return URL ││ ││ Output: ││ - download_url (string): Link to report ││ │└────────────────────────────────────────────────────────────┘Complete Example: Analytics Assistant
Section titled “Complete Example: Analytics Assistant”Here’s a complete flow for a data analytics chat assistant:
┌─────────────────────────────────────────────────────────────┐│ ││ App Setup (runs once): ││ ││ Create DataFusion Session ││ │ ││ ▼ ││ Register PostgreSQL (production database) ││ │ ││ ▼ ││ Mount CSV (reference data) ││ │ ││ ▼ ││ Store Session in Variable ││ │├─────────────────────────────────────────────────────────────┤│ ││ Chat Event Handler: ││ ││ Chat Event ││ │ ││ ├──▶ history ││ │ ││ ▼ ││ Make Agent (Claude 3.5 Sonnet) ││ │ ││ ▼ ││ Set System Prompt: "You are a data analyst..." ││ │ ││ ▼ ││ Add SQL Session (from variable) ││ │ ││ ▼ ││ Add Flow Tools: [create_chart, export_csv] ││ │ ││ ▼ ││ Add Thinking Tool ││ │ ││ ▼ ││ Invoke Agent Streaming ││ │ ││ ├── On Chunk ──▶ Push Chunk ││ │ ││ └── Done ──▶ Log completion ││ │└─────────────────────────────────────────────────────────────┘Use Cases
Section titled “Use Cases”1. Ad-Hoc Data Exploration
Section titled “1. Ad-Hoc Data Exploration”User prompts:
- “Show me sales by region for last quarter”
- “Which products have declining sales?”
- “Compare this year to last year”
2. Automated Reporting
Section titled “2. Automated Reporting”User prompts:
- “Generate a weekly sales report”
- “Create an executive summary of Q4 performance”
- “Export the top 100 customers to CSV”
3. Predictive Insights
Section titled “3. Predictive Insights”User prompts:
- “Which customers are at risk of churning?”
- “Predict next month’s revenue”
- “What factors drive customer lifetime value?“
4. Data Quality Checks
Section titled “4. Data Quality Checks”User prompts:
- “Are there any anomalies in yesterday’s data?”
- “Check for duplicate records”
- “Find missing values in the customer table”
Best Practices
Section titled “Best Practices”1. Provide Good Context
Section titled “1. Provide Good Context”Include table descriptions in your system prompt:
You have access to these tables:- orders: Order transactions (id, customer_id, amount, date)- customers: Customer info (id, name, segment, join_date)- products: Product catalog (id, name, category, price)2. Guide the Analysis Process
Section titled “2. Guide the Analysis Process”When analyzing data:1. First understand the question2. Check what data is available3. Write and execute SQL4. Summarize key findings5. Suggest visualizations or next steps3. Handle Large Results
Section titled “3. Handle Large Results”For queries that might return many rows:- Always use LIMIT unless explicitly asked for all data- Summarize results instead of showing raw data- Offer to export large datasets to files4. Enable Reasoning
Section titled “4. Enable Reasoning”Add the Thinking Tool for complex analysis:
Add Thinking Tool │ ├── Agent: (your agent) │ └── Agent ──▶ (agent with step-by-step reasoning)5. Secure Your Data
Section titled “5. Secure Your Data”- Use read-only database connections when possible
- Limit which tables the agent can access
- Log all queries for audit purposes
Combining with ML
Section titled “Combining with ML”Agents can leverage ML models you’ve trained:
Option 1: Pre-trained Model Tool
Section titled “Option 1: Pre-trained Model Tool”Create a flow that loads and runs a saved model:
Flow: predict_with_model │ ├── Input: features (array) │ ├── Load ML Model (saved model) ├── Predict │ └── Output: predictionOption 2: On-Demand Training
Section titled “Option 2: On-Demand Training”Let the agent trigger model training:
Flow: train_classifier │ ├── Input: table_name, target_column │ ├── Query Data ├── Split Dataset ├── Fit Decision Tree ├── Evaluate ├── Save Model │ └── Output: accuracy, model_pathTroubleshooting
Section titled “Troubleshooting””Agent writes invalid SQL”
Section titled “”Agent writes invalid SQL””- Include table schemas in the system prompt
- Add examples of correct queries
- Use models known for good SQL (GPT-4, Claude)
“Agent doesn’t use tools”
Section titled ““Agent doesn’t use tools””- Verify tools are properly connected
- Mention available tools in the system prompt
- Try more explicit user prompts
”Responses are slow”
Section titled “”Responses are slow””- Use streaming to show progress
- Set query timeouts
- Consider caching frequent queries
”Agent hallucinates data”
Section titled “”Agent hallucinates data””- Require the agent to always query before stating facts
- Include verification steps in the system prompt
- Log and validate SQL before execution
Next Steps
Section titled “Next Steps”Combine AI-powered analysis with:
- DataFusion & SQL – Understand the SQL capabilities
- Machine Learning – Build models the agent can use
- Data Visualization – Create charts from agent output
- AI Agents – Deep dive into agent capabilities