QuoteForge: Local AI proposals with prices you can trust

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

What I built

QuoteForge helps an independent freelancer turn a client brief into a clear proposal. Enter the service, brief, hourly rate, estimated hours, and expenses; the app calculates the quote and asks a local open-weight model to draft the words around it.

Demo

The demo runs locally. Clone the repository, then run:

docker compose up --build

After the first-run download of the Ollama runtime and llama3.2 model completes, open http://127.0.0.1:8000. The first download needs disk space and can take a while. There is no hosted demo yet.

Code

QuoteForge on GitHub

How I built it

The app uses a small Python standard-library HTTP server, a plain HTML/CSS/JavaScript interface, Docker Compose, and Ollama running the open-weight llama3.2 model. The model drafts the proposal; ordinary code calculates rate × hours + expenses, so generated text cannot alter the estimate.

The prompt asks the model not to invent experience, deliverables, dates, or guarantees. Missing details are marked for the freelancer to confirm. The draft stays in the browser until copied; the app has no account, database, analytics, or send-to-client feature.

Why open innovation matters

A client brief can include private business details. Running the model locally keeps that brief on the freelancer’s machine and avoids sending it to a hosted AI API. The model can also be changed through the OLLAMA_MODEL setting. The trade-off is that the user supplies local compute and manages the initial model download.

I kept the first version deliberately narrow: it drafts wording, while the freelancer reviews the scope and decides whether the price is right. That matches practical guidance from Shawn M. on reliability and narrow AI workflows and bestbee on gathering verifiable evidence when evaluating open-source AI tools.

Limitations

QuoteForge has not been tested with a real client workflow yet. Model output can still be wrong, so review it before sharing. The current demo is local-only and requires Docker plus enough disk space for the model.