This is a submission for the Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass
What I Built
**
ErrandQuest** is an open-source, AI-powered errand planner designed with one core philosophy: make the screen the shortest part of your day so you can get outside and touch grass.
The Problem
Most digital to-do apps keep people glued to screens—organizing, tagging, and rearranging lists—yet fail to solve real-world friction:
Which errands can be combined into a single stop?
What order should I visit shops in so I don't drive or walk in circles?
What essential items must I pack before stepping out the door?
Why did I buy ice cream first, only to have it melt while visiting the print shop?
As a result, people make 3 or 4 separate trips throughout the day, repeatedly opening their phones.
The Solution
ErrandQuest turns a scattered brain dump into a single organized outing:
10-Second Capture: Dump messy thoughts in plain English ("Buy 2L milk and veggies, print lab file, collect parcel before 5 PM, get shampoo, and withdraw cash at ATM").
Open-Source AI Extraction: A local open-weight model extracts structured tasks, identifies deadlines, warns about duplicates, and clusters errands into practical real-world stops.
Smart Outing Rules:
Prerequisites First: ATM/Bank stops are placed first so cash is ready for subsequent stops.
Urgency Prioritized: Tasks with hard deadlines (e.g., "before 5 PM") are scheduled early.
Freshness Rule: Perishable groceries (milk, ice cream, meat) are scheduled at the end of the trip so food doesn’t spoil in your bag.
Departure Checklist: Dynamically derives what to carry before leaving (cloth shopping bags, documents on USB, parcel OTP codes, ID cards, wallet).
Outdoor Execution & Carry-Forward: A high-contrast mobile checklist lets you check off stops on the go. If you end the outing early, completed tasks stay complete while unfinished errands are carried forward into your next trip.
Demo
🌐 Live Web Application: https://errandquest.onrender.com/
Mobile-Ready: Fully responsive web app designed with high contrast for outdoor sunlight visibility.
(Tip: Take a quick screenshot of the 4 steps in the app—Capture, Plan, Departure, and Live Outing—or snap a photo of you using the checklist outside to attach here!)
Code
Check out the full open-source codebase on GitHub: 👉 https://github.com/swarnimnim/errandquest
Stack:
Frontend: React 19, Vite, Mobile-first CSS
Backend: Python, FastAPI, Pydantic, SQLite
Local AI: Ollama running open-weight llama3.2:1b / qwen2.5 + Built-in deterministic NLP fallback engine
How I Built It
ErrandQuest is architected around local-first open-weight AI and deterministic planning rules:
Local Open-Weight Model via Ollama: We integrate with Ollama using lightweight models like llama3.2:1b or qwen2.5:1.5b. These models run smoothly on standard laptops without requiring a dedicated GPU or gigabytes of VRAM.
Strict Pydantic JSON Validation: Instead of unreliable free-form chatbots, ErrandQuest forces the model to respond in structured JSON adhering to a Pydantic schema (title, category, suggested_stop, urgency, deadline, prep_items, is_perishable).
Zero-Crash Hybrid Design: If a user is low on RAM or hasn't downloaded Ollama yet, ErrandQuest contains an intelligent, zero-footprint deterministic Python engine. It extracts items and plans trips in 0.01 seconds with 0 MB of extra memory, ensuring the app never stutters or crashes.
Transparent Planning Engine: Rather than asking an LLM to guess routing arbitrarily, deterministic code enforces real-world logic: prerequisite cash stops first, urgent deadlines early, and perishable goods last.
Why Does Open Innovation Matter?
Open innovation is what makes ErrandQuest viable. A closed, proprietary cloud API would be the wrong tool for this project for four critical reasons:
Intimate Daily Life Privacy:
Your errand list contains your most sensitive personal habits: prescription medications from the pharmacy, banking tasks, parcel pickup locations, and household routines. Sending that continuous stream of life data to Big Tech cloud servers for token extraction creates an unnecessary privacy risk. With open-weight local models, your personal schedule never leaves your device.
Zero Internet / Outdoor Reliability:
When you're out and about—in a park, an underground subway, or an area with spotty reception—cloud AI fails. Because ErrandQuest runs open-weight models locally on the device, it works 100% offline with zero network latency.
Zero Cost Forever:
Cloud APIs charge per token and impose monthly rate limits. Running an open-weight model on your own hardware costs zero dollars per request.
Hardware Flexibility (No Vendor Lock-in):
Open models can be swapped effortlessly. On an ultra-portable 8GB laptop, we run llama3.2:1b; on a workstation with a GPU, the exact same backend can load llama3.2:3b or mistral-7b without changing a single line of application code.
My Agent Session
I designed the complete product requirements and architecture for ErrandQuest, and partnered with an autonomous AI coding agent to turn the concept into a production-ready application.
Our session workflow:
Hardware & Resource Safety Assessment: Before writing code, the agent inspected my laptop’s available memory (identifying that an 8GB laptop with ~700MB free RAM would freeze if forced to run heavy 7B models). Together, we architected a hybrid dual-mode system: ultra-compact llama3.2:1b for local AI plus an ultra-fast deterministic fallback engine to guarantee zero system crashes.
Full-Stack Development: Scaffolded the FastAPI REST backend, SQLite relational persistence, and a mobile-first React frontend with high-contrast UI for outdoor use.
Automated Acceptance Testing: We built a dedicated verification test suite (test_suite.py
) to programmatically test all core edge cases: natural-language decomposition, duplicate detection without silent deletion, perishable reordering, and carry-forward handling for incomplete outings.
Single-Port Production Deployment: Unified static React asset serving through FastAPI to enable one-click deployment directly onto Render.
Prize Categories
Touch Grass (Grand Prize) — Open-source AI that gets people off screens and into the world.
Top comments (0)