# ScopeShield: A Local AI Scope-Creep Detector for Freelancers > Source: > Published: 2026-10-15 17:09:30+00:00 *This is a submission for the [Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle](https://dev.to/challenges/hacktoberfest-week2-2026-10-12)* ScopeShield is a local AI assistant that helps freelancers identify scope creep before agreeing to additional client work. Every freelancer knows the situation: a project starts with a clear brief, but new requests gradually appear. A client asks for another feature, extra revisions, content migration, or work that was never included in the original agreement. Freelancers must decide whether the request is already covered, needs clarification, or should be discussed as additional work. Handling this repeatedly takes time and can create uncomfortable client conversations. ScopeShield compares the original project brief with the client's new request and classifies it as: Included — the request appears to fit the original scope. Needs clarification — the brief or request is ambiguous. Likely out of scope — the request appears to involve additional work. It also provides supporting evidence, clarifying questions, recommended next steps, a change-request checklist, and a professional response the freelancer can adapt and send to the client. The intended users are freelancers, independent consultants, designers, developers, and small agencies. A potential business model could be a paid desktop tool or a subscription for professionals who manage multiple client projects. I have not yet validated willingness to pay. [https://github.com/Veerhan-glitch/ScopeShield](https://github.com/Veerhan-glitch/ScopeShield) I built ScopeShield around Gemma 4 E2B, an open-weight model that runs locally through Ollama. The application has three main parts: Frontend: HTML, CSS, and JavaScript provide fields for the original brief, the new client request, and optional context. Backend: A Python HTTP server validates the input and sends the comparison task to the local Ollama API. AI analysis: Gemma evaluates the relationship between the original scope and the new request and returns structured JSON for the interface to display. The prompt instructs the model to be fair to both parties, avoid inventing contract terms, and ask for clarification when the brief is ambiguous. It also tells the model not to provide legal advice or invent prices and timelines. I tested the application locally and worked through issues involving JSON response encoding, model request handling, and structured-output parsing. The goal is not to replace a freelancer's judgment. It is to make scope discussions easier to start and more consistent to handle. Open innovation made it possible to build and experiment with an AI-powered tool without making a hosted AI API the centre of the application. By using an open-weight model with local inference, I can develop and test ScopeShield on my own machine, inspect the application code, change the prompts, and control how client text is processed. This is especially relevant because project briefs and client communications can contain confidential business information. With the local setup, text is sent to the Ollama service on the same machine rather than intentionally transmitted to a third-party hosted AI API. Users should still verify their configuration and avoid entering confidential information without authorization. Open-weight models also make experimentation more accessible to independent developers who want to build practical products without depending entirely on a proprietary AI service. For me, open innovation is about more than access to a model. It is the ability to adapt the technology to a specific problem, understand the complete workflow, and build a useful tool around it. Gemma