Paper Reader

A productized paper recommendation tool that helps researchers decide what to read first, why it matters, and whether it is worth reproducing.

Paper Reader is built for a very practical research question:

• Which paper should I read first?
• Which one is actually worth deeper attention?
• Which one is more realistic to reproduce or track?

Instead of dumping a list of papers on the screen, it turns a topic into:

• ranked recommendation cards
• one-sentence takeaways
• clean detail pages
• comparison views for decision making
• lightweight follow-up lists

Why People Star It

• It feels like a product, not just a crawler.
• It turns paper discovery into reading decisions.
• It is easy to try with Docker and your own API key.
• It supports both lightweight public usage and local-model workflows.

Core Features

1. Daily Topic-Based Recommendations

• search by research topic such as LLM, RAG, Multimodal, or Reasoning
• get ranked paper cards instead of a raw feed
• each card shows a one-sentence takeaway and compact tags

2. Decision-Friendly Detail Pages

• recommendation conclusion
• background and goal
• summary and method notes
• reproducibility evidence
• normalized PDF download naming

3. Paper Comparison

• compare 2 to 3 papers under the same topic
• quickly judge which one deserves time first
• useful for reading, reproduction, inspiration, and related work decisions

4. My Follow-Ups

• save papers into reading list
• save papers into reproduction list
• save papers into topic candidate list
• persist follow-up items in the backend database

Quick Start

This repo is optimized for the lightest onboarding path first.

Recommended path: Docker + your own API key

You do not need Ollama, a local model download, Node.js, or Python just to try the product.

Run:

docker compose up --build

Then open:

• Frontend: http://localhost:3000
• Backend docs: http://localhost:8000/docs

Inside the app:

1. Open Model Settings
2. Choose DeepSeek, Kimi, Qwen, or another OpenAI-compatible API
3. Paste your own API key
4. Enter a topic like LLM, RAG, Reasoning, or Multimodal
5. Start using the product

This is the default public onboarding path.

Deployment Modes

Public / lightweight mode

Best for first-time users.

• Docker
• your own API key
• no Ollama required
• no local model download required

Advanced local mode

Best for users who explicitly want local inference.

Optional:

ollama pull qwen2.5:7b

Then use:

• provider: ollama
• model: qwen2.5:7b
• base URL: http://localhost:11434

Development mode

Use this only if you want to edit the code.

Frontend:

cd paper-reader-ui
npm install
npm run dev

Backend:

cd paper-reader-v1
py -3.11 -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
python -m uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

Screenshots To Add

For a stronger GitHub page, add these screenshots to the README later:

1. Homepage recommendations
2. Paper detail page
3. Compare page
4. Research toolbox / follow-up view

If you add screenshots, place them under docs/ or docs/assets/ and link them here.

Tech Stack

Frontend

• Next.js
• TypeScript

Backend

• FastAPI
• SQLite

Model Layer

• OpenAI-compatible API providers
• optional Ollama local inference

Why SQLite First

This project is intentionally kept lightweight for:

• demos
• GitHub sharing
• local product showcase
• fast first deployment

Advanced users can later replace it with an external database if they want larger-scale persistence.

See:

• paper-reader-v1/.env.example.txt

Repo Structure

paper-project/
|- README.md
|- LICENSE
|- .gitignore
|- docker-compose.yml
|- docs/
|- paper-reader-ui/
|- paper-reader-v1/
`- paper-reader skill/

Open Source Goal

This repo is not just a code dump. It is intended to show product thinking:

• ranking instead of raw listing
• decision support instead of paper collection
• follow-up workflow instead of one-time browsing
• deployability instead of environment-heavy prototypes

License

MIT