💎 Prismarine 💎

Prismarine is a research exploration tool that transforms flat lists of papers into an interactive 3D universe. By leveraging advanced AI, it visualizes semantic connections between papers.

Hackathon Defense

🛠️ Functionality (35%) - Seamless & Efficient

• Real-time 3D Rendering: Smooth, high-performance visualization of hundreds of papers using react-force-graph-3d and Three.js.
• Instant Search & Grounding: Powered by Google Search Grounding to find relevant papers in milliseconds.
• Fail-Safe Data Fetching: Robust backend handles rate limits and API failures from arXiv, PMC, and Semantic Scholar gracefully.

💡 Innovation & Creativity (25%) - Original & Feasible

• Semantic Discovery: Instead of using the reference page to find related papers, our algorithm searches for papers with similar content using Vector Embeddings, revealing hidden connections across fields.
• Accessible Research: We don't just show papers; we explain them.
  • Audio Narration: Listen to paper summaries on the go (ElevenLabs).
  • Code-Mixed Translation: Simplifies complex technical jargon into "Hinglish", "Spanglish", etc., using Snowflake to break down language barriers.

⚙️ Technical Complexity (20%) - Advanced Tech Integrated

• Multi-Model AI Pipeline:
  • Google Gemini: For vector embeddings (768d) and search grounding.
  • Snowflake: For intelligent translation and technical term preservation.
  • ElevenLabs: For high-fidelity text-to-speech.
• Complex Graph Algorithms: Real-time cosine similarity computations and K-Nearest-Neighbor linking in the backend.

🎨 UX & Design (10%) - Intuitive & Aesthetic

• Immersive Dark UI: A premium aesthetic with glassmorphism and fluid animations using framer-motion.
• Interactive 3D Space: Zoom, rotate, and click nodes to explore the citations universe. Built using D3.js and Three.js.
• Responsive Design: Works on all devices including web and mobile.

Architecture

graph TD
    subgraph Frontend [Next.js Client]
        UI[User Interface]
        ThreeD[3D Force Graph]
        State[State Management]
    end

    subgraph Backend [FastAPI Server]
        API[API Routes]
        Manager[Session Manager]
        GraphEngine[Graph Generation Engine]
        
        subgraph Services
            Auth[Auth Service]
            PaperService[Paper Service]
            Audio[Audio Service]
            Translate[Translation Service]
        end
    end

    subgraph Data [Data Persistence]
        Users[(Users)]
        Papers[(Papers & Embeddings)]
        Sessions[(Sessions)]
        Mongo[MongoDB Atlas]
    end

    subgraph AI [AI & External APIs]
        Gemini["Google Gemini API (Embeddings & Grounding)"]
        Snowflake["Snowflake Cortex (Translation & Simplification)"]
        Eleven["ElevenLabs API (Audio Generation)"]
        Sources["External Sources (arXiv, PMC, Semantic Scholar)"]
    end

    UI --> API
    ThreeD --> API
    API --> Manager
    Manager --> GraphEngine
    
    GraphEngine --> PaperService
    GraphEngine --> Services
    
    PaperService --> Gemini
    PaperService --> Sources
    
    Translate --> Snowflake
    Audio --> Eleven
    
    Manager --> Mongo
    GraphEngine --> Mongo

Loading

Key Features

1. 2D & 3D Similarity Graph

Visualize the research landscape. Papers are nodes; relationships are edges. The similarity of each node between the root is visualized through the distance between them.

Hover over a node to see the paper title and abstract. Click and hold on a node to see the the directly connected nodes and relationships. Click on a node to see the paper details.

2. Smart "Code-Mixed" Translation

Understanding a paper in a second language is hard. We use Snowflake Cortex to translate summaries while preserving technical terms in English.

• Example: "The neural network optimizes loss" -> "Ye neural network loss ko optimize karta hai" (Hindi/English mix).

3. Audio Briefs

Turn any paper abstract into a high-quality audio brief using ElevenLabs, perfect for consuming research while commuting.

Tech Stack

Frontend

• Framework: Next.js 16 (App Router)
• Styling: Tailwind CSS v4
• Visualization: react-force-graph-3d, three.js
• Animation: framer-motion
• Rendering:

Backend

• Framework: FastAPI (Python 3.10+)
• Database: MongoDB (Motor async driver)
• AI Integration: google-genai SDK, Snowflake Connector

📦 Deployment

The project is first dockerized and conatinerize using docker compose for each service such as frontend and backend. After, it then gets deployed via Coolify which is a self hosted vercel like alteunative that is run through a personal high performance server.

📄 License

MIT License. Built for CTRL+HACK+DEL 2.0