Skip to main content
This scenario is different from the four before it. There is no “chart skill” — the model uses Python directly in the code-execution sandbox: pandas to load the data, matplotlib to render it, and numpy to fit the trendline. The skill being demonstrated here is the platform itself — the fact that the model has a real Python runtime, real packages, and a way to surface PNG outputs back into the chat.

​
The prompt

Here is a CSV of weekly active users for an indie SaaS over 12 weeks.
Analyze it and produce a publication-quality PNG line chart showing the trend
with a smoothed trendline, plus a short text summary highlighting the inflection point.

week,wau
1,420
2,455
3,478
4,510
5,529
6,548
7,612
8,689
9,742
10,801
11,856
12,910

Use Python (matplotlib, pandas) in the sandbox. Save the chart to outputs and show it inline.

​
The output

SaaS user-growth analysis — analysis on the left, matplotlib chart on the right with annotated inflection point The model didn’t just produce a chart — it produced an analysis with a recommendation:
  • Headline: “Your SaaS grew from 420 to 910 weekly active users over 12 weeks — an impressive 116.7% total growth.”
  • Key findings: average weekly gain 44.5 users, starting momentum (~19 users/week for weeks 1-6), inflection point at week 6-7 (12% jump, +64 users in one week), post-inflection acceleration (~55 users/week — nearly 3x the earlier pace).
  • Interpretation: “Something shifted between Week 6 and 7 that supercharged your growth. Whether it’s product-market fit kicking in, a marketing campaign landing, or network effects activating — you’ve got momentum now.”
  • Recommendation: “Double down on what worked around Week 6-7. That’s your secret sauce.”
The chart on the right is a real matplotlib PNG — line plot with markers, a fitted trendline, and two highlighted yellow regions calling out the starting-momentum phase and the inflection point around Week 6.

​
Why it works

Three platform pieces, no skill needed:
  1. Code execution — full Python sandbox with pandas, matplotlib, numpy, scipy, etc. pre-installed.
  2. File mounts — the chart is saved to /mnt/user-data/outputs/ and surfaced back to the chat as an inline image.
  3. The model’s own reasoning — given the data, it picked an inflection point and built a story around it. That’s not a skill, that’s the model doing what LLMs are good at, with real numbers in front of it.
This is also where the platform shines compared to “AI artifact” demos: the chart is a file you can download, not an embedded image you can’t reuse. Right-click → save → drop into a deck. Next: Roasthaus landing page (live in browser) →