🦥 We built a pixel-art office where five cats do your data analysis while a sloth naps 🐱

Fiona Khoo — Sun, 25 Oct 2026 16:27:33 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady

Every dataset starts the same way: df.head(), count the missing values, find the duplicates, draw a histogram, repeat. So we handed that work to a team of cats.

Purrrfessorrr Slothhh is a local-first data analysis assistant. Upload a CSV, TSV, or Excel file, and a very sleepy sloth professor delegates the work to five specialist cats. Your data never leaves your machine.

What I Built

🕵️ Detective Cat finds outliers, duplicates, and type problems
🎨 Artist Cat draws the charts
📝 Scribe Cat explains the findings in plain English
💻 Coder Cat writes Python and R code for every analysis
🧹 Janitor Cat suggests how to clean the data

Ask follow-up questions in chat, then export everything as a ZIP or a Jupyter notebook.

Demo

Try it in two minutes. No account and no API key needed:

git clone https://github.com/ashtonckj/Purrrfessorrr-Slothhh.git
cd Purrrfessorrr-Slothhh
.\run.bat        # macOS / Linux: ./run.sh

Code

GitHub: https://github.com/ashtonckj/Purrrfessorrr-Slothhh

How I Built It

Python + FastAPI, vanilla JavaScript, and optional Gemma 3 running locally through Ollama.

The key decision was to keep calculation and conversation separate:

Python computes every number. The analysis is a plain, documented pipeline, so the same file always gives the same report.
Gemma only explains. It sees a summary of the computed results, not your raw rows. A guard then removes any sentence containing a number that isn't in the real data.
No model? Still works. Without Ollama, the full analysis and rule-based chat keep running.

Why Does Open Innovation Matter?

A dataset is often the most sensitive thing on your laptop. Open-weight models let us build AI-assisted analysis that needs no hosted API and sends nothing to a third party. We enforced it in code: the app refuses to talk to any model that isn't on localhost.

The Lazy Payoff

The repetitive first pass of exploratory analysis (profiling, quality checks, starter charts, boilerplate code) happens in one guided workflow, so you spend your time on what the data means. We haven't benchmarked the time saved, so we won't claim a number.

Prize Categories

Best Use of Gemma (featured): Gemma 3 runs locally via Ollama to write the morning digest, grounded strictly on the model's own output, with a deterministic offline fallback so it never hard-fails.

👥 Built by three humans (and five fictional cats)

Ashton — Github: @ashtonckj · Dev: @ashtonckj
Fiona — Github: @fionakhl · Dev: @fionakhl
Jocelyn — Github: @JocelynKoh3 · Dev: @moltenice3

🎨 Third-party Credits

While the application itself was developed by us, some of the illustrations and icons were sourced from artists and creative communities across the web. 🐈✨

A huge thank you to the creators whose artwork helped give this project its playful personality! 💛

We've documented the sources of these assets in my GitHub repository under ASSETS.md, where you can explore the original artwork and its attribution details.

All credit for the third-party illustrations belongs to their respective creators.

The sloth is slow. The workflow doesn't have to be. 🦥💤

DEV Community: Fiona Khoo