BookFlip Tutor: A Local AI Coach for Smarter Book Reselling

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

What I Built

BookFlip Tutor is a local-first learning tool for students, freelancers, and first-time book resellers who want to know whether a used book is worth buying before they spend money.

A user can choose a seeded example or paste a marketplace listing. BookFlip Tutor then helps turn the listing into a resale decision by showing:

• Buy price and inbound shipping
• Expected sale price and marketplace fees
• Estimated profit and return on investment
• Risk flags for missing ISBN, unclear edition, shipping cost, and thin margins
• A plain-language explanation of what to verify next

The product is designed for a small, specific side hustle: helping a beginner avoid bad textbook purchases and learn the reasoning behind a good sourcing decision. A paid version could charge for saved sourcing history, guided textbook niches, batch listing review, or a private local workflow for tutors and small resellers.

The financial calculation is deterministic. The AI is not trusted to invent prices or perform arithmetic.

Demo

Open index.html from the repository, or serve the project folder with any static file server.

Code

Repository: github.com/HarshDev-07/DevProj-2

The main implementation is intentionally small and dependency-free:

• index.html contains the product workflow.
• styles.css contains the responsive interface.
• app.js contains seeded listings, Ollama integration, validation, deterministic calculations, risk flags, quiz behavior, and local saved history.
• README.md documents setup, architecture, privacy, and limitations.
• LICENSE provides the MIT license.

How I Built It

BookFlip Tutor is a browser app with an optional local AI path through Ollama.

The user can configure a lightweight local model such as qwen2.5:3b or another compatible open-weight model. The app sends the pasted listing to Ollama's local /api/generate endpoint with:

• stream: false for a simple browser workflow
• format: "json" for structured output
• Low temperature for more consistent extraction
• A prompt that explicitly says not to guess missing numbers

The expected response contains structured fields such as:

{
  "title": "Campbell Biology",
  "edition": "12th edition",
  "condition": "Very good",
  "buyPrice": 38,
  "shipping": 6,
  "expectedSalePrice": 96,
  "marketplaceFee": 14.4,
  "missing": []
}

The app validates numeric fields before using them, exposes the extracted values for manual correction, and keeps the existing calculator available if Ollama is not installed or running.

The architecture is:

flowchart LR
  A[Marketplace listing text] --> B[Ollama open-weight model]
  B --> C[Validated facts and uncertainty]
  C --> D[Manual review and correction]
  D --> E[Deterministic profit calculator]
  E --> F[Estimate, risk flags, and lesson]
  F --> G[Local browser history]

To run Ollama locally:

ollama pull qwen2.5:3b
ollama serve

Then open the app, click Check connection, paste a listing, and click Extract with Ollama.

Why Does Open Innovation Matter?

The local model is what makes this product practical as a small side hustle rather than another usage-metered API wrapper.

Privacy: Book listings, sourcing notes, and saved opportunities can stay on the user's laptop instead of being sent to a third-party inference service.

Cost: A lightweight open-weight model can run without a per-request bill. That matters when a beginner is checking low-cost books where a hosted API charge could consume the entire margin.

Choice: Users can replace the model with another Ollama-compatible model, choose a smaller model for an older laptop, or later tune the workflow for textbooks, exam-prep books, or a regional marketplace.

Transparency: The user can see and correct extracted fields before they affect the estimate. The app keeps financial rules in ordinary application code, so the model cannot silently change the profit calculation.

Offline potential: Once the model is downloaded, the extraction workflow can work without an internet connection. The current app does not require a hosted backend for its core seeded examples, calculations, lessons, or saved history.

AI Use Disclosure

I used AI assistance during development for implementation support, code drafting, interface iteration, and documentation. The project also uses Ollama as an optional runtime for an open-weight model. The product keeps model output reviewable and does not treat AI-generated extraction as a guaranteed fact.

Prize Categories

Best Use of GitHub Copilot

BookFlip Tutor was built with GitHub Copilot assistance for implementation, debugging, interface iteration, and documentation. Copilot helped turn the product concept into a working browser demo while the project remains open-source and understandable as a small codebase.

Categories Not Claimed

I am not claiming the Render, TabPFN, Tinker, Arduino, DigitalOcean, Gemma, Backboard, ElevenLabs, Entire, Mastra, MongoDB Atlas, Sentry, SerpApi, .tech, Temporal, or Tiger Data categories because the current implementation does not use those technologies.