ScopeSmith — an AI proposal generator that runs on a $0 local model

Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle Submission 💸

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

ScopeSmith turns a messy client brief into a client-ready proposal in under a minute. Paste a rambling WhatsApp export, a half-formed Upwork message, a rambling email — you get back a structured proposal: scope of work, concrete deliverables, an explicit out-of-scope list, the top scope-creep risks with countermeasures, and freelancer-friendly payment terms (deposit, milestones, revision policy).

It's for freelancers, because freelancers lose money two ways: slow proposals and vague scope. A client sends a rambling message on a Tuesday night; the freelancer spends an evening turning it into a proposal, forgets to define what's out of scope, and three weeks later is doing unpaid "small tweaks" at midnight. One bad project costs more than a year of tooling.

Why would someone pay for this? The math is brutal and simple:

No hosted demo yet — but here's a real end-to-end run, generated by the actual local model ( Qwen2.5-1.5B-Instruct via Ollama) from a messy bakery-owner brief:

Input: "hey! saw your portfolio, love your style. so i run a small bakery in austin called Crumb & Craft and we desperately need a website… budget is around $2500-3000… launch before thanksgiving rush, so like 4-5 weeks?"

Deliverables: Homepage, Menu page, Online ordering system, Contact/order form

Out of Scope: Logo and photos (already provided), Domain/hosting setup (already handled)

pip install -r requirements.txt ollama pull qwen2.5:1.5b python -m scopesmith --brief examples/brief_webdesign.txt --out proposal.md Enter fullscreen mode Exit fullscreen mode Code Repository: [public link coming soon — code will be open-sourced on GitHub]

scopesmith/ providers.py # LLMProvider ABC + ollama / openai-compatible / mock backends prompts.py # the four stage prompts + proposal renderer pipeline.py # extract → scope → red flags → terms orchestration cli.py # `python -m scopesmith` entry point examples/ # three realistic messy client briefs (web, mobile app, copywriting) outputs/ # generated sample proposals Enter fullscreen mode Exit fullscreen mode How I Built It The key architectural decision: four small, single-job prompts beat one giant prompt — extract → scope → red flags → terms. That matters because ScopeSmith runs on small local models. The default qwen2.5:1.5b (~1 GB) follows tight instructions far better than sprawling mega-prompts, and the four-stage pipeline keeps each step's output structured enough for the next.

The model layer is a swappable provider interface: ollama (default, local open-weight models — qwen2.5:1.5b, llama3.1, gemma2 ), openai (any OpenAI-compatible endpoint: LM Studio, vLLM, llama.cpp server), and a deterministic mock provider that keeps the demo runnable with no model at all.

Three reasons, all load-bearing — not decorative:

None — not built on partner technology. Entering the overall prize only.

Built October 14, 2026 for the DEV Hacktoberfest Open-Source AI Challenge, Week 2 (Side Hustle). AI belongs to everyone — including the freelancer trying to get paid fairly for every hour they work.