ScopeNote: a local AI second look before a freelance quote
This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle . What I Built ScopeNote is a local AI second look before a freelance quote. It turns a client brief into requested scope, questions to ask, scope risks, and a draft reply. Consider this fictional request: a fiv
Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle Submission 馃捀
This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle.
ScopeNote is a local AI second look before a freelance quote. It turns a client brief into requested scope, questions to ask, scope risks, and a draft reply.
Consider this fictional request: a five-page bakery website by Friday, online ordering, a new logo, an $800 budget, and unlimited revisions. Before discussing a price, a freelancer needs to clarify what is actually being requested. ScopeNote makes that intake review repeatable without automatically accepting work or sending anything to a client.
The intended customers are solo designers, developers, and small studios. A possible paid offering would combine local installation, studio-specific intake checklists, and maintenance. This is a prototype: there are no claimed paying customers, revenue, or validated willingness-to-pay results.
Click Show recorded example to see a real local-model response to the fictional bakery brief. The public site is explicitly a recorded preview, not a hosted inference service. The repository explains how to run the app locally for a different brief.
The recorded CPU run took 25.38 seconds. That is one observation on the development machine, not a general performance benchmark. Its output flags unlimited revisions and the mismatch between the requested features and budget. It also illustrates the limitations: the reply is wordy, and the model does not follow every formatting preference.
The application code is MIT licensed. Model weights are not included in the repository.
The server binds to the local machine and checks input length, content type, Host, Origin, and a request header. It processes one generation at a time. Responses are displayed as plain text, and saving a Markdown copy requires an explicit click. The app does not automatically save client briefs or send replies.
Six automated boundary tests passed, covering valid requests, invalid input, cross-origin requests, Host restrictions, request guards, and preventing arbitrary file serving. The real local inference and deployed recorded preview were also checked separately.
An open-weight model and inspectable runtime make local processing practical. A freelancer can keep briefs on their own machine, inspect the prompt, and adapt the workflow without a per-request model-provider bill. Hardware, electricity, and setup still have costs.
The model can miss details, repeat itself, and introduce assumptions. ScopeNote is an intake aid requiring human review, not a pricing engine or contract generator. Prompt instructions are not a guarantee against misleading output, so the model has no tools or automatic side effects.
AI assistance disclosure: This prototype and this write-up were created with AI coding and writing assistance. The demonstration output comes from an actual local Phi model run; it is not a simulated response.