FixFlow：以開源 AI 將零散維修投訴轉化為可跟進工單

FixFlow: Turning Messy Maintenance Complaints into Actionable Tickets with Open-Source AI

This is a submission for the "Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle" (https://dev.to/challenges/hacktoberfest-week2-2026-10-12).

What I Built

FixFlow is an AI-powered maintenance request manager designed for small hostels, PG accommodations, and property managers.

Residents often report problems informally: a leaking tap, a broken light, a faulty fan, or a water supply issue. These complaints can get lost in chats, lack important details, or remain unresolved because nobody knows who is responsible.

FixFlow uses the open-weight Gemma model to transform natural-language complaints into structured maintenance tickets. It extracts the issue category, summary, location, missing information, and suggested priority. When essential details are missing, it can ask follow-up questions instead of making assumptions.

Managers can use a dashboard to track requests, assign staff, update statuses, identify possible duplicate tickets, and monitor maintenance activity.

The goal is simple: make maintenance easier to report, easier to manage, and harder to forget.

Who could benefit? Small hostel owners, PG operators, student accommodation managers, and property maintenance teams that need a lightweight alternative to complicated property management systems.

Screenshot

Code

GitHub repository: https://github.com/abirami-prabhakar/flowfix

The project is designed around a simple Python and Streamlit application, with SQLite for persistence and Gemma through Ollama for local inference.

How I Built It

I chose a deliberately lightweight architecture:

• Gemma + Ollama: Extracts structured information from unstructured maintenance complaints using an open-weight model running locally.
• Python: Handles the application logic, input processing, validation, and ticket workflow.
• Streamlit: Provides the resident request form and manager dashboard.
• SQLite: Stores tickets and their status history.
• Pydantic: Validates AI-generated structured output before it is accepted by the application.
• Sentry: Supports error monitoring and performance tracing, where configured.
• Render: A potential deployment target, subject to the requirements of running Ollama and providing persistent storage.

The AI is not treated as an unquestionable decision-maker. Its output must be validated, missing details must remain visible, and potentially urgent complaints should be flagged for human review.

I also designed the workflow around predictable failure handling. Malformed model responses, inference failures, and duplicate-looking requests should not silently produce unreliable records. A deterministic demo mode can make testing possible without requiring a live model for every test.

Why Does Open Innovation Matter?

Open innovation makes the core AI functionality more accessible and gives developers control over how models run and how data moves through an application.

With Gemma running locally through Ollama, FixFlow can process complaints without sending their contents to a hosted AI inference API. This is particularly useful for accommodation providers handling room information and potentially sensitive resident reports.

It also makes experimentation easier. Developers can change the model, inspect its behavior, adjust prompts, and test structured extraction against realistic complaints without being locked into one proprietary inference service.

Open-weight models do not automatically guarantee privacy or reliability. Those still depend on how the application is deployed, how data is stored, and how the model's output is validated. But they give smaller teams more control over those decisions.

That is what makes this approach interesting to me: useful AI should not require every small organization to build its workflow around a closed API.

Prize Categories

Open-source AI / Gemma, where eligible under the challenge's official category rules.