The Grass Receipt -- Turn your offline walks into printable physical receipts.

T Abishek — Sat, 10 Oct 2026 19:56:50 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Touch Grass

The Problem: The Screen-Time Paradox

When you spend your days wrestling with complex code, algorithms, and computational physics, your brain hits a wall. Staring at a display longer yields zero marginal productivity. The best debugger is walking away from the screen.

However, standard habit trackers feel like chores. I wanted to build something lighthearted, fast, and completely offline that rewards offline time with physical art rather than another screen notification.

What I Built: The Grass Receipt

The Grass Receipt is a local app that turns your offline walks and screen-free breaks into a physical, vintage store receipt.

You input your micro-stats (minutes outside, burnout level, and screen-free time). TabPFN instantly evaluates your metrics to calculate a "Mental Reset Score," and a local Gemma 2 model formats a witty, monospaced "receipt" of your outdoor experience. You hit print, grab your glossy paper, and tape it straight to your wall.

How I Built It

TabPFN (Tabular Foundation Model): Evaluates the user's offline habit data to generate the mental reset classification probability in milliseconds without tedious manual training loops.
Gemma 2 (gemma2:2b via Ollama): Acts as the local text generation engine, turning raw scores into aesthetic, structured store receipts.
Sentry Agent Tracing: Tracks transaction performance across the TabPFN calculation and LLM inference pipeline.
DevRelay Integration: Session telemetry tracking the development and building process.

Why Open Source & Local AI Matter

An app designed to disconnect you from screens shouldn't rely on cloud subscription tokens or active Wi-Fi connections when you're sitting on a bench outside. By utilizing a zero-shot tabular model like TabPFN and a fast local weight like Gemma 2, everything runs locally on machine hardware in milliseconds.

The Handover

Take a 30-minute walk, drop your stats into the app, print the receipt, and turn off your computer.

Prize Categories Entered

Best Use of TabPFN: Predicting mental reset scores from tabular offline habit data.
Best Use of Gemma: Generating localized offline receipt text via Ollama.
Best Use of Sentry Agent Tracing: Execution telemetry tracking.
Best Use of DevRelay: Session documentation.

My Friend Was Guessing Hostel Mess Portions, So I Built Him an Offline AI Macro Agent

T Abishek — Mon, 05 Oct 2026 06:58:20 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

If you’ve ever lived in a freshers' hostel, you know the daily struggle: you start the semester with massive fitness goals, buy a brand-new tub of whey protein, hit the gym religiously, and then walk straight into the campus mess hall to face whatever unidentifiable, liquid-gold oily curry the kitchen decided to whip up today.

The problem? Commercial fitness apps like MyFitnessPal are fundamentally built for Western grocery stores or standardised restaurant chains. Typing "hostel mess dal fry" or "random cafeteria paneer butter masala" into a generic macro tracker yields wildly inconsistent estimates-one database entry says a bowl is 100 calories, another says 450 calories, and none of them tells you how much you should actually scoop onto your tray to hit your specific targets without overshooting calories.

Watching him stare soulfully at a tray of mystery gravy every afternoon, trying to calculate mental math while the lunch queue backed up behind him, was equal parts inspiring and tragic. I decided to fix it.

Demo

Code

Abishekcps / hostelmacro

How I Built It

I built this as a local Streamlit application running on Windows. Instead of relying on a giant cloud API, I wanted the brain of this thing sitting right on his laptop, so it actually works when he needs it.

Core Intelligence (Open Weights): I used Google's open-weight Gemma 2 (gemma2:2b) running locally via Ollama. It takes the messy hostel menu, the web context, and his macro goals, and acts as the reasoning engine to build the actual plate strategy.

Web Tooling: I used the DuckDuckGo Search Python package for a keyless, zero-auth live web context. It pulls real-time nutritional profiles for specific Indian dishes before the LLM generates an answer, so it doesn't hallucinate calories.

Tracing & Performance: I integrated Sentry Agent Tracing via their Python SDK. It monitors the execution timeline, creating transaction spans for both the web tool search latency and the local LLM generation time, so I can see exactly where the bottleneck is if the app feels slow.

Why Does Open Innovation Matter?

Why didn't I wrap a closed cloud API in a website? Two words: Cost and Connectivity.

When you are checking your macros three times a day, 30 days a month, API token bills add up fast for a broke college student. By using an open-weight model like Gemma 2, this app costs absolutely $0 to run forever.

Secondly, hostel Wi-Fi can be notoriously spotty, especially during peak hours. Building this with local inference means the core intelligence works even when the internet drops completely. Furthermore, his personal health goals and daily dietary logs remain 100% private on his own hardware. Open source means anyone reading this can clone the repo, run ollama pull gemma2:2b, and have a fully functional AI assistant running on their machine tonight without setting up credit cards or secret API keys.

Prize Categories:

Best Use of Gemma: Running Google's open-weight gemma2:2b locally via Ollama as the core reasoning engine for macro calculations and meal planning.

Best Use of Sentry Agent Tracing: Full telemetry integration tracking tool execution spans, error logging, and latency profiling across the local LLM inference and web context tools.

DEV Community: T Abishek

The Grass Receipt -- Turn your offline walks into printable physical receipts.

The Problem: The Screen-Time Paradox

What I Built: The Grass Receipt

How I Built It

Why Open Source & Local AI Matter

The Handover

Prize Categories Entered

My Friend Was Guessing Hostel Mess Portions, So I Built Him an Offline AI Macro Agent

What I Built

Demo

Code

Abishekcps / hostelmacro

How I Built It

Why Does Open Innovation Matter?

Prize Categories: