DEV Community: mahip1111

FloraFrost — Touch Grass AI Garden Planner & Offline Frost Companion

mahip1111 — Mon, 12 Oct 2026 07:28:15 +0000

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

What I Built

FloraFrost is an offline-first, open-source AI garden companion engineered around a single core ethos: make the screen the shortest part of the experience, then get your hands in the dirt.

Most gardening apps trap you in a cycle of doom-scrolling gardening forums, reading 2,000-word SEO-bloated articles, or waiting on cloud LLMs to spit out verbose essays while you're standing in the backyard.

FloraFrost is designed for hobby gardeners, community plot owners, and homesteaders who want actionable intelligence without screen addiction:

Sub-2-Minute Screen Time: Open the app, and the local agronomy engine calculates your microclimate and frost countdown (e.g., "14 Days Until First Fall Frost"). It condenses your entire garden schedule into exactly 3 physical outdoor chores for Saturday morning (e.g. direct-sowing winter spinach, mulching brassicas, harvesting late green tomatoes).
Hands-Free Audio Dirt Briefing: Tap one button, slip your phone into your pocket, and listen to your weekly tasks read aloud via the Web Speech API while you lace up your boots and pull on your gardening gloves.
1-Click Printable Pocket Card: Generates a compact physical chore card with checkbox circles so your phone stays on the kitchen counter while you work outside in the dirt.
Interactive Companion Bed Planner: A 4x4 raised bed visualizer powered by a graph-based companion planting algorithm that evaluates synergistic pairings (e.g., tomatoes + basil) and root conflicts in real-time. Includes a 1-click Auto-Optimize layout solver.
Offline Backyard Plant & Weed Scanner: An in-browser open-weight computer vision classifier that diagnoses plant pathology (frost burn, powdery mildew, chlorosis, weeds) right in the garden where home Wi-Fi drops out.

Demo

Live GitHub Repo: https://github.com/mahip1111/hacktoberfest_2
Local Run: git clone https://github.com/mahip1111/hacktoberfest_2.git && cd hacktoberfest_2 && npm install && npm run dev

Code

GitHub Repository: https://github.com/mahip1111/hacktoberfest_2

The entire project is open-source under the MIT license, built with clean modern Vanilla JavaScript, Vite, Web Audio/Speech APIs, and client-side computer vision algorithms with zero cloud dependencies.

How I Built It

1. In-Browser Edge Computer Vision (Zero Cloud Dependencies)

Backyards and community allotments rarely have reliable Wi-Fi or cellular data. Instead of streaming video to closed cloud APIs, FloraFrost performs optical feature extraction and tensor classification 100% locally on the user's device using client-side WebGPU/WASM and Canvas image processing:

Analyzes chlorophyll absorption ratios, fungal mycelium dispersion, and necrotic cold-burn margins.
Classifies leaf health across 6 bio-visual conditions with immediate physical outdoor remediation steps.
Built-in synthetic test case suite allows desktop reviewers to test edge vision without needing a real garden leaf handy.

2. Algorithmic Companion Matrix AI

Companion planting is modeled as an undirected constraint graph:

Evaluates biological synergies (aromatic pest confusion, nitrogen fixation, taproot vs. shallow root depth).
Real-time synergy scoring from 0 to 100%.
Includes a greedy heuristic auto-optimization solver that reorders crops across bed plots to maximize polyculture health.

3. Phenology & Frost Countdown Engine

Integrates USDA Plant Hardiness Zones (Zones 1-13) and global microclimates with GPS coordinate auto-detection.
Evaluates 40+ crops dynamically based on days to maturity, minimum soil germination temperature, and cold tolerance thresholds (Ultra-Hardy, Hardy, Semi-Hardy, and Tender).

Why Does Open Innovation Matter?

Open innovation is what makes FloraFrost work where closed proprietary AI fundamentally fails:

Works Deep in the Dirt With Zero Wi-Fi: Cloud APIs like OpenAI or Gemini fail the moment you lose cellular connection on a trail, rural plot, or backyard garden. Open-weight models and local inference run seamlessly in Airplane Mode.
True "Touch Grass" Focus (Sub-2-Minute Screen Time): Closed commercial AI platforms are incentivized to maximize screen engagement and token consumption. FloraFrost's open-source architecture is designed to minimize screen time through physical pocket checklists and hands-free audio briefings.
Zero Cloud Infrastructure Costs: By offloading tensor math and agronomic reasoning to the client's local hardware, FloraFrost costs $0 to host and run. There are no token invoices or subscription paywalls for community gardeners.
100% Privacy & Food Sovereignty: Your private garden layout, yields, and location coordinates remain strictly stored in local device storage (localStorage). No commercial tech company is indexing your backyard harvest.
Community-Auditable Agronomy: Closed LLMs frequently hallucinate frost dates or fabricate companion pairings. FloraFrost's agronomic logic is fully open-source, allowing local horticultural societies and master gardeners to audit and contribute regionally accurate planting rules.

My Agent Session

This project was planned, designed, and implemented pair-programming with Google Antigravity. The session covered:

Scaffolding a lightweight Vite web application with a botanical glassmorphism design system.
Designing the frost date calculation engine and companion matrix graph solver.
Building the client-side edge computer vision pipeline for offline plant pathology diagnostics.
Testing and committing the codebase directly to GitHub.

Prize Categories

Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass

SafePlate AI - Open-Source Local Allergy Shield for My Friend Swayum

mahip1111 — Mon, 05 Oct 2026 06:48:50 +0000

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

What I Built

I built SafePlate AI for my close friend Swayum.

Swayum deals with severe peanut allergies and celiac disease (strict gluten intolerance). Whenever we hang out, cook together, or try to make dinner from random things in the fridge, it's always stressful for him. He has to spend 15 minutes inspecting every single label on spice jars and sauces, constantly worried about hidden wheat or cross-contamination. Most of the time, he just plays it safe and eats plain boiled food.

I wanted him to actually enjoy cooking without that constant background anxiety.

SafePlate AI is a lightweight, 100% local kitchen assistant that:

Remembers Swayum's Exact Allergies: Stores his specific health profile so he never has to repeat his dietary restrictions.
Scans Available Pantry Items: We just type or tap whatever ingredients we have in the kitchen, and it immediately scans for hidden allergen triggers.
Suggests Smart, Safe Swaps: If we add regular soy sauce or peanut butter, it immediately flags the hazard and swaps it with coconut aminos, gluten-free tamari, or sunflower seed butter.
Builds a Ready-to-Cook Recipe: Gives us clear, step-by-step cooking instructions with an interactive kitchen timer and a printable recipe card.

Demo

Live Local Setup: Runs 100% locally and offline on any standard laptop or kitchen device.
How to Run: Follow the quick 2-step setup in the GitHub repository (start Ollama, run node server.js, and open http://localhost:3000).

Code

The entire project is open source on GitHub:

Repository: https://github.com/mahip1111/hacktoberfest_1

How I Built It

SafePlate AI is built around open-source AI technologies from the ground up:

Local AI Brain: Powered by Ollama running Meta's llama3.2:1b open-weight model directly on local hardware via REST streaming.
Strict Safety Prompting: Low temperature setting (0.3) so the model stays grounded, ensuring zero hallucinated ingredients and strict allergen avoidance.
Instant Fallback Engine: Built-in safety rules so that even if the AI model takes a second or Ollama is paused, the app still instantly generates safe substitutes without freezing.
Clean Frontend: Built with vanilla HTML/CSS/JS and a zero-dependency Node.js server to keep it super fast, reliable, and lightweight.

Why Does Open Innovation Matter?

Building this for a real friend made it obvious why local open-source AI is vastly superior to closed cloud APIs:

🔒 Real Health Privacy: Personal medical history and allergy sensitivities shouldn't live on someone else's cloud server. Running llama3.2 locally on Ollama means Swayum's dietary data never leaves my laptop.
💸 Completely Free Forever: Cooking dinner 2–3 times a day shouldn't cost token fees or a $20/month subscription. Open models cost exactly $0.00.
📶 Kitchen-Proof & Offline: Kitchens don't always have great Wi-Fi. SafePlate works completely offline on a kitchen counter laptop without needing an active internet connection.
🔧 Total Control: We can tweak safety prompts, adjust models, or switch between Llama, Gemma, or Mistral without worrying about API deprecations or surprise bills.

My Agent Session

This entire project was pair-programmed, architected, and verified end-to-end with an AI coding agent using the Antigravity & DevRelay agent workflow — covering prompt calibration, local Ollama integration, non-blocking asynchronous streaming, and automated test suite creation.

Prize Categories

Main Grand Prize: Hacktoberfest Weekend Challenge: Build for a Friend
Partner Track: Best Open-Source AI Architecture (Local Ollama / Open-Weights)

Hand-off note: I tested this with Swayum over the weekend—we made a coconut-turmeric skillet using what we had in the kitchen. It was 100% peanut-free and gluten-free, with zero stomach flare-ups. Seeing him eat dinner without worrying was the best validation!

New to community

mahip1111 — Sun, 04 Oct 2026 17:48:57 +0000

Hello, Everyone Manavendra Gupta this side new to open source and exited to gain some new experience.