How to Talk to Nova Effectively

How to Talk to Nova Effectively

Nova understands natural language, but knowing how to communicate clearly will help you get better results faster. This guide covers prompting strategies, common patterns, and tips for productive AI interactions.

The Basics: What Nova Understands

Nova can interpret:

• Goals: What you want to accomplish
• Context: Your data, files, and previous work
• Constraints: Resource limits, time requirements, specific methods
• Preferences: Output formats, visualization styles, level of detail

Prompting Patterns

Pattern 1: Goal-First Prompts

Start with what you want to achieve, then add details.

Good:

Analyze my sales data to find seasonal trends.
Use the sales_2024.csv file in my stash.
Create visualizations showing monthly patterns.

Less Effective:

I have a CSV file called sales_2024.csv and I was wondering
if maybe you could look at it and see if there are any
patterns or something?

Pattern 2: Specify Your Data

Tell Nova exactly where your data lives.

Good:

Using the protein.pdb file in stash://myusername/simulations/,
run a GROMACS energy minimization.

Also Good:

Connect to my Snowflake warehouse and query the
ANALYTICS.CUSTOMER_EVENTS table for the last 30 days.

Pattern 3: Request Specific Outputs

Be clear about what you want back.

Good:

Generate a summary report with:
- Key statistics (mean, median, std dev)
- A histogram of the distribution
- Top 10 outliers with explanations
Save everything to my stash.

Pattern 4: Iterative Refinement

Build on previous results with follow-up questions.

# First prompt
Analyze gene expression data from rnaseq_results.csv

# Follow-up
Focus on genes with log2 fold change > 2

# Another follow-up
Create a volcano plot highlighting the significant genes

Common Scenarios

Data Analysis

@nova What can you tell me about the dataset in
stash://myusername/data/experiment_results.csv?
I want to understand the distribution and identify any anomalies.

Running Scientific Applications

@nova Run a variant calling pipeline on my FASTQ files
in stash://myusername/genomics/samples/.
Use the GRCh38 reference genome.

Code Generation

@nova Write a Python script that:
1. Loads data from my Snowflake connection
2. Performs time series forecasting
3. Saves predictions to my stash

Job Management

@nova What's the status of my running jobs?
Show me the logs for job 12345.

Tips for Better Results

Be Specific About Methods

Instead of: “Analyze this data” Try: “Perform a principal component analysis and show the first 3 components”

Mention File Formats

Instead of: “Save the results” Try: “Save results as a CSV file and create a PDF report”

Set Expectations

Instead of: “Make it fast” Try: “Use a SMALL compute node, this is a quick test”

Ask for Explanations

Add: “Explain your methodology” or “Walk me through the results”

What Nova Can’t Do

• Access external systems not connected via Data Connectors
• Modify production databases without explicit confirmation
• Run indefinitely long computations (jobs have time limits)
• Access other users’ private data

Troubleshooting

Nova seems confused:

• Break your request into smaller steps
• Be more explicit about file paths and data sources
• Check that your data connectors are properly configured

Results aren’t what you expected:

• Ask Nova to explain its approach
• Provide more context about your domain
• Specify the exact method or algorithm you want

Jobs are failing:

• Ask Nova to check the job logs
• Verify your input files exist and are accessible
• Try a smaller test case first

Next: Code Generation with Nova | Best Practices