Expert Streamlit Dashboard

Expert Streamlit Dashboard Patterns

Overview

This skill provides professional mental models and design patterns for building production-grade Streamlit dashboards. It emphasizes a "Data-First, Minimalist" aesthetic (avoiding "vibes" coding or excessive emojis) and rigorous performance optimization for SQL-backed applications.

When to Use

• Executive Reporting: Building dashboards for decision-makers who need clear signals, not noise.
• Large Datasets: Handling millions of rows efficiently without crashing the browser or server.
• Interactive Analytics: Complex filtering scenarios where performance is critical.
• Internal Tools: Replacing spreadsheet workflows with robust data apps.

The Mental Shift: "Vibe Coding" vs. Expert Engineering

Core Design Patterns

1. The "Data-First" Minimalist UI

Professional dashboards respect the user's attention.

Pattern:

• No Emojis in Headers: Use clean typography.
• Key Metrics (KPIs) Top-Left: The most valuable real estate.
• Consistent Color Palette: Use st.set_page_config and config.toml to enforce brand colors, not random selections.
• Whitespace: Use st.container and st.columns to create breathing room.

2. Performance: The Caching Strategy

Streamlit re-runs the entire script on every interaction. Without caching, your app is unusable.

Pattern:

• Data Cache (@st.cache_data): For serializable data (Pandas DataFrames, JSON).
  • Crucial: Set ttl (Time To Live) to avoid stale data (e.g., ttl=300 for 5 mins).
  • Crucial: Set max_entries to prevent memory leaks.
• Resource Cache (@st.cache_resource): For database connections, ML models.

@st.cache_resource
def get_db_connection():
    return mysql.connector.connect(...)

@st.cache_data(ttl=300, show_spinner=False)
def get_sales_data(start_date, end_date):
    # ... executes optimized SQL ...
    return df

3. SQL Optimization: Push-Down Computation

Never load a massive table into Pandas to filter it.

Pattern:

• Filter in SQL: Pass dashboard filters (start_date, region) directly into the WHERE clause.
• Aggregate in SQL: If the user wants "Sales by Month", do GROUP BY in SQL, not Pandas.
• Limit Rows: Always have a safety valve (LIMIT 10000) or pagination.

Bad: SELECT * FROM sales -> df[df['date'] > start] Good: SELECT * FROM sales WHERE date > :start

4. Intuitive Filtering System

Filters should be powerful but unobtrusive.

Pattern:

• Sidebar First: Place all global controls in st.sidebar.
• Dynamic Defaults: Set default filter values to the "most useful" view (e.g., "Last 7 Days" instead of "All Time").
• Form Batching: For expensive queries, wrap filters in st.form so the app only re-runs when the user clicks "Apply".

with st.sidebar.form(key='filters'):
    date_range = st.date_input("Date Range", [])
    submit_button = st.form_submit_button(label='Update Dashboard')

if submit_button:
    # Run query

5. Advanced State Management

Don't let users lose their place.

Pattern:

• URL Sync: Sync key filter states to URL parameters using st.query_params so links can be shared.
• Session State: Use st.session_state to store intermediate calculations or user preferences that shouldn't trigger a full data reload.

Expert Checklist

Before deploying your dashboard:

• Load Test: Does it handle 1M rows? (If not, optimize SQL).
• Cache Hit Rate: Are expensive queries actually being cached? (Check debug logs).
• Visual Hierarchy: clear distinction between KPIs, Charts, and Data Tables.
• Mobile Responsiveness: Do columns stack correctly on smaller screens?
• Error Handling: Are st.error or st.warning used instead of raw Python tracebacks?

Common Anti-Patterns

❌ "The Wall of Text": Dumping a raw dataframe (st.dataframe(df)) without specific intent. ❌ "Blocking Queries": Running long SQL queries in the main thread without a spinner or cache. ❌ "Widget Explosion": Putting 20 filters on the screen at once. Use "Advanced Filters" expanders. ❌ "Hardcoded Secrets": Never put DB credentials in code. Use .streamlit/secrets.toml.