Prompt Guide

Ready-to-use prompts for data analysis, machine learning, visualization, and more. Run them in Zerve now.

Popular Prompts

Data Analysis

Exploratory Data Analysis

Perform comprehensive EDA on a CSV dataset with summary statistics, distributions, and correlations.

Load the dataset from the uploaded CSV file and perform a full exploratory data analysis. Include: (1) shape and data types, (2) missing value summary, (3) descriptive statistics for all numeric columns, (4) distribution plots for key variables, (5) a correlation heatmap, and (6) top 3 actionable insights from the data.

Try Now

Machine Learning

Classification Model

Train and evaluate a binary classification model with feature importance analysis.

Build a binary classification model to predict the target column. Steps: (1) split data 80/20 with stratification, (2) train a Random Forest and XGBoost model, (3) compare using accuracy, precision, recall, F1, and AUC-ROC, (4) plot ROC curves for both models, (5) show top 10 feature importances as a bar chart, and (6) recommend which model to deploy and why.

Try Now

Visualization

Interactive Dashboard

Create a multi-chart interactive dashboard with filters and drill-downs.

Build an interactive Plotly dashboard with the dataset. Include: (1) a KPI summary row showing total revenue, active users, and growth rate, (2) a line chart for trend over time with a date range selector, (3) a bar chart comparing categories with click-to-filter, (4) a scatter plot showing the relationship between two key metrics with size encoding, (5) a geographic map if location data exists, and (6) a dark theme matching Zerve's brand colors.

Try Now

NLP & Text

Sentiment Analysis

Analyze text data for sentiment, emotions, and key themes.

Perform sentiment analysis on the text data. Steps: (1) clean and preprocess the text (remove stopwords, lowercase, lemmatize), (2) classify each entry as positive, negative, or neutral with a confidence score, (3) extract the top 20 most frequent themes/topics using TF-IDF, (4) create a word cloud for positive and negative sentiments separately, (5) plot sentiment distribution over time if dates are available, and (6) summarize the top 3 complaints and top 3 praises.

Try Now

23 prompts

Data Analysis

3 prompts

Exploratory Data Analysis

Perform comprehensive EDA on a CSV dataset with summary statistics, distributions, and correlations.

Load the dataset from the uploaded CSV file and perform a full exploratory data analysis. Include: (1) shape and data types, (2) missing value summary, (3) descriptive statistics for all numeric columns, (4) distribution plots for key variables, (5) a correlation heatmap, and (6) top 3 actionable insights from the data.

Try Now

Data Cleaning Pipeline

Build a reusable pipeline to clean messy data: handle missing values, outliers, and type conversions.

Create a data cleaning pipeline for my dataset. Steps: (1) identify and report missing values per column, (2) impute missing numerics with median and categoricals with mode, (3) detect and cap outliers using IQR, (4) convert date columns to datetime, (5) standardize column names to snake_case, and (6) output a summary of all transformations applied.

Try Now

Cohort Retention Analysis

Analyze user retention by monthly cohorts with a heatmap visualization.

Using the transactions data, perform a cohort analysis: (1) group users by their first purchase month, (2) calculate monthly retention rates for each cohort, (3) create a retention heatmap with percentages, (4) highlight the cohort with the best and worst 3-month retention, and (5) suggest 2 strategies to improve early retention.

Try Now

Machine Learning

3 prompts

Classification Model

Train and evaluate a binary classification model with feature importance analysis.

Build a binary classification model to predict the target column. Steps: (1) split data 80/20 with stratification, (2) train a Random Forest and XGBoost model, (3) compare using accuracy, precision, recall, F1, and AUC-ROC, (4) plot ROC curves for both models, (5) show top 10 feature importances as a bar chart, and (6) recommend which model to deploy and why.

Try Now

Time Series Forecast

Forecast future values using ARIMA or Prophet with confidence intervals.

Forecast the next 30 days of the target metric using time series analysis. Steps: (1) plot the historical trend and check for seasonality, (2) run an ADF test for stationarity, (3) fit both ARIMA and Prophet models, (4) plot forecasts with 95% confidence intervals, (5) calculate MAPE and RMSE for each model on a holdout set, and (6) provide the final forecast table with daily predictions.

Try Now

Hyperparameter Tuning

Optimize model performance with grid search or Bayesian optimization.

Optimize the hyperparameters for my XGBoost model. Steps: (1) define a search space for learning_rate, max_depth, n_estimators, subsample, and colsample_bytree, (2) run Bayesian optimization with 50 trials using Optuna, (3) use 5-fold cross-validation with AUC-ROC as the metric, (4) plot the optimization history and parameter importances, (5) train the final model with best params, and (6) compare performance before and after tuning.

Try Now

Visualization

3 prompts

Interactive Dashboard

Create a multi-chart interactive dashboard with filters and drill-downs.

Build an interactive Plotly dashboard with the dataset. Include: (1) a KPI summary row showing total revenue, active users, and growth rate, (2) a line chart for trend over time with a date range selector, (3) a bar chart comparing categories with click-to-filter, (4) a scatter plot showing the relationship between two key metrics with size encoding, (5) a geographic map if location data exists, and (6) a dark theme matching Zerve's brand colors.

Try Now

Chart Storytelling

Generate publication-ready charts with annotations and narrative context.

Create 5 publication-ready charts that tell a story from this data. For each chart: (1) choose the most appropriate chart type, (2) add a clear title and subtitle explaining the insight, (3) annotate key data points or inflection moments, (4) use a consistent color palette, (5) include source attribution, and (6) write a 2-sentence narrative caption for each chart suitable for a report.

Try Now

Geospatial Analysis

Map data geographically with choropleth maps and spatial clustering.

Create a geospatial analysis from the location data. Steps: (1) plot all data points on an interactive Folium map, (2) create a choropleth by region/state showing the key metric, (3) run DBSCAN spatial clustering to find hotspots, (4) calculate distance-based statistics, (5) add a layer toggle for different metrics, and (6) export the map as an HTML file.

Try Now

Data Engineering

3 prompts

ETL Pipeline Builder

Design an extract-transform-load pipeline with validation and logging.

Design a production-ready ETL pipeline. Steps: (1) extract data from the specified source (CSV/API/database), (2) validate schema and data quality with Great Expectations-style checks, (3) transform: rename columns, join tables, aggregate metrics, (4) add row-level logging for failures, (5) load into the target destination with upsert logic, and (6) generate a run summary with row counts, duration, and any quality warnings.

Try Now

SQL Query Optimizer

Analyze and optimize slow SQL queries with index suggestions.

Optimize this SQL query for performance. Steps: (1) explain the current execution plan, (2) identify bottlenecks (full table scans, missing indexes, suboptimal joins), (3) rewrite the query with optimizations, (4) suggest indexes to create with their expected impact, (5) provide the optimized EXPLAIN output, and (6) estimate the performance improvement as a percentage.

Try Now

Database Schema Design

Design a normalized database schema from business requirements.

Design a database schema for the following use case. Steps: (1) identify all entities and relationships from the requirements, (2) create a normalized schema (3NF) with proper data types, (3) define primary keys, foreign keys, and unique constraints, (4) add indexes for common query patterns, (5) generate CREATE TABLE statements, and (6) draw an ER diagram showing all relationships.

Try Now

Automation

2 prompts

Data Workflow Automation

Automate repetitive data tasks with scheduled execution and notifications.

Automate this data workflow. Steps: (1) fetch fresh data from the API endpoint every morning at 8 AM, (2) run the cleaning and transformation pipeline, (3) update the dashboard datasets, (4) check all KPIs against their targets, (5) send a Slack notification with a summary if any KPI is below target, and (6) log all run metadata (timestamp, duration, status, rows processed) to a tracking table.

Try Now

API Data Connector

Pull data from REST APIs with pagination, rate limiting, and error handling.

Build a robust API data connector. Steps: (1) authenticate with the API using the provided credentials, (2) implement pagination to fetch all records, (3) add exponential backoff retry logic for rate limits and errors, (4) parse the JSON response and flatten nested structures, (5) validate the output schema and data types, and (6) save to a DataFrame with a timestamp column for when each record was fetched.

Try Now

Reporting

3 prompts

Automated Report Generator

Generate a formatted weekly/monthly report from live data sources.

Build an automated weekly report. Include: (1) an executive summary with 3 key takeaways, (2) week-over-week metrics comparison with arrows for direction, (3) a top performers table (top 10 by the primary metric), (4) an anomaly section flagging any metric that moved more than 2 standard deviations, (5) a forecast for the next week based on recent trends, and (6) format everything as a clean HTML report with inline styles.

Try Now

A/B Test Analysis

Evaluate experiment results with statistical significance and power analysis.

Analyze this A/B test data. Steps: (1) calculate conversion rates for control and treatment groups, (2) run a chi-squared test and report the p-value, (3) compute the confidence interval for the lift, (4) perform a power analysis to check if sample size was sufficient, (5) segment results by key dimensions (device, region, cohort), and (6) make a clear ship/no-ship recommendation with reasoning.

Try Now

Anomaly Detection Report

Detect and report anomalies in time series metrics with root cause suggestions.

Run anomaly detection on the time series metrics. Steps: (1) apply Isolation Forest and Z-score methods, (2) flag all anomalous data points with severity levels, (3) plot the time series with anomalies highlighted in red, (4) for each anomaly, check correlated metrics for possible root causes, (5) generate a summary table with date, metric, expected vs actual, and severity, and (6) suggest monitoring thresholds for future alerting.

Try Now

NLP & Text

3 prompts

Sentiment Analysis

Analyze text data for sentiment, emotions, and key themes.

Perform sentiment analysis on the text data. Steps: (1) clean and preprocess the text (remove stopwords, lowercase, lemmatize), (2) classify each entry as positive, negative, or neutral with a confidence score, (3) extract the top 20 most frequent themes/topics using TF-IDF, (4) create a word cloud for positive and negative sentiments separately, (5) plot sentiment distribution over time if dates are available, and (6) summarize the top 3 complaints and top 3 praises.

Try Now

Document Summarizer

Summarize long documents into key points with configurable detail levels.

Summarize the provided documents. For each document: (1) extract the 5 most important key points, (2) generate a 1-paragraph executive summary, (3) identify all named entities (people, companies, dates, amounts), (4) list action items or recommendations mentioned, (5) assign topic tags from a predefined taxonomy, and (6) create a comparison table if multiple documents are provided.

Try Now

Topic Modeling

Discover hidden topics in a text corpus using LDA or BERTopic.

Run topic modeling on the text corpus. Steps: (1) preprocess text with tokenization, stopword removal, and lemmatization, (2) fit an LDA model with 5-10 topics and display the top 10 words per topic, (3) also fit a BERTopic model for comparison, (4) create an interactive topic visualization (pyLDAvis or similar), (5) assign the dominant topic to each document, and (6) name each topic based on its keywords and provide a 1-sentence description.

Try Now

Research

1 prompt

Interactive Executive Report

Build an interactive report for executive stakeholders with drill-down charts and KPI tracking.

Create an interactive executive report. Steps: (1) define 5 top-level KPIs with sparkline trends and RAG status indicators, (2) build drill-down charts — clicking a KPI reveals the underlying breakdown by region, product, or time period, (3) add a date range selector and comparison toggle (vs prior period / vs target), (4) include a risks and opportunities section with auto-flagged items based on threshold rules, (5) generate a one-paragraph AI summary of the current performance, and (6) export the full report as a self-contained HTML file with embedded data suitable for email distribution to C-suite.

Try Now

Data Applications

2 prompts

Streamlit Data App

Build a Streamlit application with interactive widgets, data visualization, and user inputs.

Build a Streamlit data application. Steps: (1) create a sidebar with file upload (CSV/Excel), date range filters, and a multi-select for key dimensions, (2) display a summary metrics row at the top with st.metric showing value, delta, and delta color, (3) add an interactive Plotly chart section with tabs for trend, comparison, and distribution views, (4) include a filterable data table with download button using st.dataframe, (5) add a text input where users can ask natural-language questions about the data and display AI-generated answers, and (6) style the app with a custom theme matching dark mode with purple accent colors.

Try Now

Gradio ML Demo App

Build a Gradio application to demo a machine learning model with interactive inputs and outputs.

Build a Gradio demo application for a machine learning model. Steps: (1) create a tabbed interface with 'Single Prediction' and 'Batch Prediction' tabs, (2) on the single tab, add input components matching the model's features — sliders for numerics, dropdowns for categoricals, and file upload for images if applicable, (3) display prediction results with a confidence gauge, class probabilities bar chart, and a plain-English explanation, (4) on the batch tab, accept a CSV upload and return predictions as a downloadable table with a summary distribution chart, (5) add an 'About' tab with model card info (accuracy, training data, limitations), and (6) deploy with gr.Blocks layout, custom CSS for dark theme, and flagging enabled so users can report bad predictions.

Try Now

User Voices

Zerve in Their Words

Real stories from people using Zerve every day

Zerve feels different because it was built for data scientists from the ground up. It understands context, adapts as you work, and lets you iterate without crashes or lost work.

Eduardo Ordax

Gen AI Lead at AWS

Zerve finally feels like the cursor for data scientists. It understands that schemas live in your files, not in your head.

Lior Alexander

Founder/CEO at AlphaSignal

I gave it a churn analysis task and it pulled data, handled missing values, ran segmentation, and scheduled the workflow automatically using context from my past work.

Eric Vyacheslav

AI Engineer at Stealth

We were working with a 4.3 billion-row dataset, and a high-end notebook instance could only handle 18 million-rows at a time without going out of memory due to the nature of preprocessing to be done on the data. That meant running the same call 24 times in a row. With Zerve, we could run all 24 in parallel. That shift alone saved us a huge amount of time and let us move faster on model development.

Arindam Bose

Head of Data Science at Sun King

The Zerve agent has transformed our data science workflow. Backlogs are vanishing and we're experimenting faster than ever.

Mike Smith

Founder/CEO at Ascendr.org

Zerve feels different because it was built for data scientists from the ground up. It understands context, adapts as you work, and lets you iterate without crashes or lost work.

Eduardo Ordax

Gen AI Lead at AWS

Zerve finally feels like the cursor for data scientists. It understands that schemas live in your files, not in your head.

Lior Alexander

Founder/CEO at AlphaSignal

I gave it a churn analysis task and it pulled data, handled missing values, ran segmentation, and scheduled the workflow automatically using context from my past work.

Eric Vyacheslav

AI Engineer at Stealth

Arindam Bose

Head of Data Science at Sun King

The Zerve agent has transformed our data science workflow. Backlogs are vanishing and we're experimenting faster than ever.

Mike Smith

Founder/CEO at Ascendr.org

As we advance our AI and Data standardization initiatives, Zerve provides the robust foundation we need to scale with confidence and be a force multiplier for meaningful outcomes.

Richard Springer

Director of Data at Cubic

The Cursor for Data Science is here.

Sandipan Bhaumik

Data & AI Lead at DataBricks

Zerve represents a step-change for data science development.

Rob Hickey

Former EVP of Engineering at DataRobot

Zerve's Agentic code assistance is easily among the best I've seen

Conor Nolan

Technologist at HPE

I used Zerve AI, and it honestly felt like a cheat code for data science.

Muhammad Tayyab Alqan

I use other AI when the answer already exists and I need to surface it. I use Zerve when the answer doesn't yet exist so I need to build it.

Warren Paull

Applied AI Consultant at Commercial Outcomes

As we advance our AI and Data standardization initiatives, Zerve provides the robust foundation we need to scale with confidence and be a force multiplier for meaningful outcomes.

Richard Springer

Director of Data at Cubic

The Cursor for Data Science is here.

Sandipan Bhaumik

Data & AI Lead at DataBricks

Zerve represents a step-change for data science development.

Rob Hickey

Former EVP of Engineering at DataRobot

Zerve's Agentic code assistance is easily among the best I've seen

Conor Nolan

Technologist at HPE

I used Zerve AI, and it honestly felt like a cheat code for data science.

Muhammad Tayyab Alqan

I use other AI when the answer already exists and I need to surface it. I use Zerve when the answer doesn't yet exist so I need to build it.

Warren Paull

Applied AI Consultant at Commercial Outcomes

Decision-grade data work

Explore, analyze and deploy your first project in minutes