I built Greenstep – An Open-Source App That Picks Your Next Walk

A developer built Greenstep, an open-source app that recommends when, where, and what kind of walk to take, using a three-part pipeline of TabPFN for hourly predictions, SerpApi for nearby place suggestions, and Gemma 3 4B for generating the final action. The app is built with FastAPI, server-rendered Jinja templates, HTMX, and SQLite, and runs Gemma locally via Ollama in development or through Google AI Studio's endpoint in deployment, switched by a single environment variable.

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass https://dev.to/challenges/hacktoberfest-week1-2026-10-05 I built Greenstep , an app that tells you when to go outside. Most apps in this space do the same thing: they remind you to take a break, go for a walk, get some air. The advice is right. It just doesn't help. You already know you should go outside. The problem is that at 6 PM, after a full day, nothing tells you this is the moment, and this is the place, and this is what you'll actually do when you get there. So you stay in. Greenstep answers those three questions: The output looks like this: Walk through Lodhi Garden in Delhi. 25 minutes. The evening is mild and your energy is low, so a slow loop fits. First step: put on your walking shoes. Fallback: sit on a bench outside for ten minutes. That's the whole thing. No streaks, no badges, no dashboard with nine things to check. It sits on a second monitor and asks for one decision a day. It's for anyone who knows a walk would help but needs someone else to make the call. FastAPI with server-rendered Jinja templates and HTMX, SQLite for the history, and a three-part pipeline behind it. TabPFN reads a 150-row CSV of past evenings and predicts the best hour for today. This is the part I care about most. TabPFN is a tabular foundation model, not a language model — it's built for exactly this kind of problem, small structured data with no training loop. I log a row for each hour I could have gone out, with sleep, screen time, mood, weather, and whether I actually went. TabPFN looks at today's numbers and returns a probability per hour. SerpApi takes the predicted hour and fetches a real place nearby. Parks in the late afternoon, cafes in the evening, walking loops if the weather is good. The suggestion names the place. Gemma 3 4B writes the final action. During development it runs locally through Ollama. For the deployed version it runs on Google AI Studio's Gemma endpoint, so anyone can try it without installing anything. The switch between local and cloud is one environment variable: