AppPlan.md

AppPlan

Explain Art to Me is an AI-powered web application designed to make art more accessible and understandable for people without formal art knowledge. The app addresses the common problem faced by students, casual art enthusiasts, museum visitors, and online users who encounter famous artworks but struggle to understand their deeper meaning because existing explanations are often too technical and filled with art jargon. Its goal is to bridge the gap between art and everyday users by explaining artworks in simple, friendly, and emotionally relatable language, helping users appreciate not just what they see, but also the story and emotions behind the artwork.

Users can either upload an image of an artwork or enter the name of a painting such as The Starry Night, and optionally choose an explanation mode like Beginner or Expert. Using an LLM API with optional RAG for retrieving artwork information, the app identifies the artwork and generates a structured explanation as output. The expected output includes five key sections: a brief introduction to the artist, a description of the scene depicted, the underlying emotions or symbolism, why the artwork is famous, and the personal or historical context in which it was created. The model acts as an empathetic art guide, focusing on storytelling, emotional connection, and clarity rather than technical art criticism.

App idea

## App idea

- Use case: [What problem is the app solving?]

- Intended user: [Who is the app for?]

- User input: [What will the user type, upload, or choose?]

- Model output: [What should the model return?]

- System prompt requirements: [If this is an LLM or RAG app, what role should the model play, what rules should it follow, and what format should it return? If this is not an LLM app, write not applicable.]

- Approach: [LLM API, RAG, linear regression, logistic regression, random forest, XGBoost, CNN, MobileNetV2, or something else]

- Important source material or dataset: [Documents, examples, spreadsheet, image dataset, or none]

The first goal is a basic working app. For LLM and RAG apps, Cowork/Codex should create a clearly labeled system prompt in the code that your team can edit later. Once the first version runs locally, iterate with Cowork/Codex to improve the interface, source material, system prompt, examples, and output format.

Product notes

• App name: [Working name]
• Tone and style: [Professional, playful, clinical, concise, etc.]

Design ideas

Treat these design choices as requirements, not loose suggestions.

• Colors: [Any preferred colors, brand colors, or colors to avoid]
• Fonts: [Simple, formal, playful, editorial, dashboard-like, etc.]
• Layout: [Single page, sidebar, form plus results panel, chat-style, dashboard, etc.]
• Visual tone: [Minimal, warm, clinical, premium, utilitarian, playful, etc.]

Notes to Cowork/Codex

If the app requires interactive front-end behavior or machine learning models, use React or FastAPI as necessary. If the app uses an LLM or RAG, make the system prompt explicit in the code instead of burying it inside a long request string.