{"slug": "pitchcraft-local-a-freelance-proposal-generator-that-runs-locally", "title": "PitchCraft Local — A Freelance Proposal Generator That Runs Locally", "summary": "A developer built PitchCraft Local, an open-source Python command-line tool that turns a client brief into an editable freelance proposal with itemized pricing using a locally run Qwen 2.5 3B model via Ollama. The pipeline keeps financial calculations outside the model, using Decimal arithmetic and deterministic pricing logic, with bounded inference attempts, offline tests, loopback-only runtime communication, and safeguards against prompt injection from client text. The developer says the MVP is designed for resource-constrained hardware such as an Intel i3 with 8 GB RAM and notes the Qwen Research License covering the model weights differs from the project's MIT license.", "body_md": "*This is a submission for the [Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle](https://dev.to/challenges/hacktoberfest-week2-2026-10-12).*\n\n**PitchCraft Local** is a command-line tool that turns a client's brief into an editable freelance proposal with itemized pricing — powered by a local open-weight AI model.\n\nFreelancers often spend time translating vague client requirements into a clear scope of work, estimating hours, calculating costs, and preparing professional proposals. PitchCraft Local helps streamline that process without requiring a paid AI subscription or cloud-based LLM API.\n\nHere's what it does:\n\nThe target users are independent freelancers, solo developers, and small service businesses that repeatedly prepare project quotations.\n\nThe goal isn't to replace a freelancer's judgment. It's to reduce repetitive drafting and arithmetic so they can focus on scope clarification, realistic estimates, and client communication.\n\nThe repository includes screenshots of the CLI running locally, including model readiness checks, generated proposals, and tests.\n\nThe project runs from a terminal and is designed for resource-constrained hardware, including an Intel i3 machine with 8 GB RAM. CPU inference may be slow, depending on the brief and machine.\n\n**GitHub repository:** [GhananilShirpurkar/pitchcraft-local](https://github.com/GhananilShirpurkar/pitchcraft-local)\n\nThe project is written in Python and includes a CLI, configurable service catalogue, example client briefs, offline tests, and an evaluation runner.\n\nStart with the repository README for installation instructions and a complete walkthrough.\n\nI built PitchCraft Local around **Python, Ollama, Qwen 2.5 3B, Pydantic, and deterministic pricing logic**.\n\nI deliberately chose a small local model because my development machine has an Intel i3 processor and 8 GB of RAM. Instead of relying on a large hosted model, I wanted to explore how useful a constrained, locally running model could be for a specific business workflow.\n\nThe pipeline looks like this:\n\n`Decimal` arithmetic.\nA key design decision was to keep financial calculations outside the model. An LLM can help interpret a brief, but it shouldn't be trusted to perform authoritative pricing arithmetic.\n\nI also added bounded inference attempts, offline tests, loopback-only runtime communication, and safeguards against treating client-provided text as trusted instructions.\n\nThe project is an MVP, not a production-ready commercial product. Its generated scope, estimates, and wording still require human review.\n\nOpen innovation made it possible to experiment with an AI-powered business tool without depending on a proprietary inference API.\n\nUsing an open-weight model through Ollama gave me control over the runtime, the inference configuration, and how data moves through the application. It also let me build and test the workflow on modest hardware instead of designing around an expensive hosted model.\n\nMore importantly, this project explores a practical engineering question: **How much of a useful AI product can be built with a small local model when the surrounding software is designed carefully?**\n\nThe answer isn't that the model can do everything. It's that the model can handle language-heavy drafting while conventional code handles validation, pricing, file safety, and predictable business rules.\n\nThat separation makes the application easier to test, inspect, and improve.\n\nOne important caveat: the repository's MIT license covers the project code, not the model weights. The selected Qwen 2.5 3B model uses a separate Qwen Research License, which must be reviewed for the intended use.\n\nI haven't included a DevRelay agent-session link in this submission. The repository and its implementation are available above.\n\n**Primary fit:** Side Hustle — a practical tool aimed at helping freelancers reduce repetitive proposal preparation.", "url": "https://wpnews.pro/news/pitchcraft-local-a-freelance-proposal-generator-that-runs-locally", "canonical_source": "https://dev.to/ghananilshirpurkar/pitchcraft-local-a-freelance-proposal-generator-that-runs-locally-3o84", "published_at": "2026-10-17 10:21:00+00:00", "updated_at": "2026-10-17 10:34:18.848678+00:00", "lang": "en", "topics": ["ai-tools", "large-language-models", "ai-products", "developer-tools"], "entities": ["PitchCraft Local", "Ollama", "Qwen 2.5 3B", "Pydantic", "Python", "GhananilShirpurkar", "Hacktoberfest"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/pitchcraft-local-a-freelance-proposal-generator-that-runs-locally", "markdown": "https://wpnews.pro/news/pitchcraft-local-a-freelance-proposal-generator-that-runs-locally.md", "text": "https://wpnews.pro/news/pitchcraft-local-a-freelance-proposal-generator-that-runs-locally.txt", "jsonld": "https://wpnews.pro/news/pitchcraft-local-a-freelance-proposal-generator-that-runs-locally.jsonld"}}