Streamlit Cheatsheet: Build Dashboards Faster with AI   

Over the past year I migrated my team's dashboards from GUI-based BI tools to a fully code-driven workflow with Python + Streamlit, using AI coding assistants like Claude Code and Cursor to speed up iteration.

Right now I'm migrating dashboards off Looker (Google Data Studio) and aggregating information from various tools into Streamlit dashboards. The reason is simple: it saves time to build a POC with AI, and then I can continue the updates myself. No waiting on a BI license, no drag-and-drop ceiling — just Python.

Sample prompt

The example below gives OK results for a POC — enough to see something on screen, not enough to ship:

create me Streamlit dashboard about

compare multiple sites:
* section onsite analysis
* serp analysis
    * ranking
    * search results - google, bing, duckduckgo
* keyword comparison
    - home page: title, headings etc
* pricing
* features

files:
* dashboard.py - UI
* data_processor.py - data preprocessing
* visualization.py - data plots and figures
* scraper.py - data collector
* config.yaml - configuration

add more if needed to make it better

But to get better results, smaller steps are needed. Instead of one big prompt, I iterate file by file with concrete instructions, then move to the next one. For example:

dashboard.py

• concise layout
• no tabs
• sections
  • onsite — basic info about titles, pages
  • serp — compare serp data

So far the results are mixed. Still experimenting.

What actually drives good results

In my experience, the output depends mostly on:

• The initial idea — I prefer to outline the idea in my head first, then on paper. The clearer the shape before prompting, the better.
• How the idea is shared with the model — smaller steps beat one giant prompt. Deterministic tests and validation keep the agent honest.
• Inspecting the code yourself — always verify accuracy rather than trusting the first output.

Dashboard design principles

Good code is only half the battle. A dashboard that actually works needs to respect a few principles:

• Show only important information — minimize or avoid scrolling so key metrics land in the first glance.
• Organize info in grouped levels — use containers, columns, and tabs to create hierarchy rather than dumping everything on one page.
• Use color deliberately — make changes and indicators visible with color, but reserve it for what actually needs attention.
• Adopt a narrative approach — tell a story with your data, not just raw numbers and graphs.
• Make it self-explanatory — design so viewers need minimal additional questions to understand what they're seeing.
• Structure as a master dashboard plus details — personally, I prefer one master dashboard with top-level info, supported by a few others for deeper detail.

The full cheatsheet is based on cheat-sheet.streamlit.app. If you're already fluent in Python, this should get you from idea to working dashboard in minutes rather than hours.

Checking Examples and Asking the Model to Update

Sometimes I browse the examples from streamlit.io/gallery?category=favorites to get inspiration, then ask the model to update my dashboard based on what I see. Below are some Streamlit dashboard examples from the web that I've collected — useful as reference points when prompting an AI agent.

So far the results from this approach are mixed — some prompts produce a solid layout out of the gate, others need several rounds of iteration. Still experimenting.

Note: that often dashboards can disappear. That's why I'm creating a local library of good dashboard examples. For this I'm using Obsidian as a master organizer for:

• dashboards
• Jupyter notebooks
• Python scripts
• datasets

Cheatsheet

Setup

Install and import Streamlit

pip install streamlit

Install streamlit

import streamlit as st

Import convention

a = st.sidebar.radio('Choose:',[1,2])

Add widgets to sidebar (just add it after st.sidebar)

'_This_ is some __Markdown__' a=3 'dataframe:', data

Magic commands implicitly st.write()

streamlit --help streamlit run your_script.py streamlit hello streamlit config show streamlit cache clear streamlit docs streamlit --version

Command line

pip uninstall streamlit pip install streamlit-nightly --upgrade

Pre-release features

Display text

Display text

st.text('Fixed width text')

Fixed width text

st.markdown('_Markdown_') # optional kwarg unsafe_allow_html = True

Markdown

st.caption('Balloons. Hundreds of them...')

Caption

st.latex(r''' e^{i\pi} + 1 = 0 ''')

LaTeX

st.write('Most objects') # df, err, func, keras!

Write most objects

st.write(['st', 'is <', 3])

Write a list

st.title('My title')

Title

st.header('My header')

Header

st.subheader('My sub')

Subheader

st.code('for i in range(8): foo()')

Code block

Display data

Display data

st.dataframe(my_dataframe)

DataFrame as an interactive table

st.table(data.iloc[0:10])

Static table

st.json({'foo':'bar','fu':'ba'})

JSON object

st.metric(label='Temp', value='273 K', delta='1.2 K')

Metric with delta

Display media

Display media

st.image('./header.png')

Image

st.audio(data)

Audio

st.video(data)

Video

Columns

Columns

col1, col2 = st.columns(2) col1.write('Column 1') col2.write('Column 2')

Create two columns and write into them

col1, col2, col3 = st.columns([3,1,1])

Three columns with different widths (col1 is wider)

with col1: st.write('This is column 1')

Using 'with' notation

Tabs

Tabs

tab1, tab2 = st.tabs(['Tab 1', 'Tab2']) tab1.write('this is tab 1') tab2.write('this is tab 2')

Insert containers separated into tabs

with tab1: st.radio('Select one:', [1, 2])

Tabs using 'with' notation

Control flow

Control flow

st.stop()

Stop execution immediately

st.experimental_rerun()

Rerun script immediately

with st.form(key='my_form'): username = st.text_input('Username') password = st.text_input('Password') st.form_submit_button('Login')

Group multiple widgets in a form

Personalize apps for users

Personalize apps for users

if st.user.email == '[email protected]': display_jane_content() elif st.user.email == '[email protected]': display_adam_content() else: st.write('Please contact us to get access!')

Show different content based on the user's email address

Display interactive widgets

Display interactive widgets

st.button('Hit me')

Button

st.data_editor('Edit data', data)

Data editor

st.checkbox('Check me out')

Checkbox

st.radio('Pick one:', ['nose','ear'])

Radio

st.selectbox('Select', [1,2,3])

Selectbox

st.multiselect('Multiselect', [1,2,3])

Multiselect

st.slider('Slide me', min_value=0, max_value=10)

Slider

st.select_slider('Slide to select', options=[1,'2'])

Select slider

st.text_input('Enter some text')

Text input

st.number_input('Enter a number')

Number input

st.text_area('Area for textual entry')

Text area

st.date_input('Date input')

Date input

st.time_input('Time entry')

Time input

st.file_uploader('File uploader')

File uploader

st.download_button('On the dl', data)

Download button

st.camera_input('一二三,茄子!')

Camera input

st.color_picker('Pick a color')

Color picker

for i in range(int(st.number_input('Num:'))): foo()

Use widgets' returned values in variables

st.slider('Pick a number', 0, 100, disabled=True)

Disable widgets to remove interactivity

Build chat-based apps

Build chat-based apps

with st.chat_message('user'): st.write('Hello 👋') st.line_chart(np.random.randn(30, 3))

Insert a chat message container

st.chat_input('Say something')

Display a chat input widget

Mutate data

Mutate data

element = st.dataframe(df1) element.add_rows(df2)

Add rows to a dataframe after showing it

element = st.line_chart(df1) element.add_rows(df2)

Add rows to a chart after showing it

Display code

Display code

with st.echo(): st.write('Code will be executed and printed')

Echo code

Placeholders, help, and options

Placeholders, help, and options

element = st.empty() element.line_chart(...) element.text_input(...) # Replaces previous.

Replace any single element

elements = st.container() elements.line_chart(...) st.write('Hello') elements.text_input(...) # Appears above 'Hello'.

Insert out of order

st.help(pandas.DataFrame)

Help

st.get_option(key) st.set_option(key, value)

Get and set options

st.set_page_config(layout='wide')

Page config

st.experimental_show(objects)

Show objects

st.experimental_get_query_params() st.experimental_set_query_params(**params)

Query params

Connect to data sources

Connect to data sources

st.experimental_connection('pets_db', type='sql')

Connection with type

conn = st.experimental_connection('sql')

SQL connection

conn = st.experimental_connection('snowpark')

Snowpark connection

class MyConnection(ExperimentalBaseConnection[myconn.MyConnection]): def _connect(self, **kwargs) -> MyConnection: return myconn.connect(**self._secrets, **kwargs) def query(self, query): return self._instance.query(query)

Create your own connection class

Optimize performance

Optimize performance

@st.cache_data def foo(bar): # Do something expensive and return data return data d1 = foo(ref1) d2 = foo(ref1) # cached, d1 == d2 d3 = foo(ref2) # different arg, executes foo.clear() st.cache_data.clear()

Cache data objects (cached by value)

@st.cache_resource def foo(bar): # Create and return a non-data object return session s1 = foo(ref1) s2 = foo(ref1) # cached, s1 == s2 s3 = foo(ref2) # different arg, executes foo.clear() st.cache_resource.clear()

Cache global resources (cached by reference)

@st.cache def foo(bar): # Do something expensive in here... return data d1 = foo(ref1) d2 = foo(ref1) # cached, d1 == d2 d3 = foo(ref2)

Deprecated caching (@st.cache)

Display progress and status

Display progress and status

with st.spinner(text='In progress'): time.sleep(3) st.success('Done')

Show a spinner during a process

bar = st.progress(50) time.sleep(3) bar.progress(100)

Show and update progress bar

st.balloons()

Balloons

st.snow()

Snow

st.toast('Mr Stay-Puft')

Toast

st.error('Error message')

Error message

st.warning('Warning message')

Warning message

st.info('Info message')

Info message

st.success('Success message')

Success message

st.exception(e)

Exception