ContextLynx

ContextLynx is a next-generation Personal Knowledge Management (PKM) tool designed to eliminate the need for manual categorization or tagging of notes. By leveraging advanced natural language processing techniques, ContextLynx organizes and connects your notes seamlessly through a dynamic knowledge graph. The application automatically detects relationships between your ideas, notes, and research topics, making your knowledge organization smarter and more intuitive. You can try out the service at contextlynx.com/get-started.

Prototype Features

• Knowledge Graph, with "real edges" represented in black and predicted edges in red. It is possible to click on any node to see related notes. Visualize your notes and their relationships in a knowledge graph, helping you explore topics holistically.

• Semantic Note Taking: Automatically detects connections between your notes using local models for Named Entity Recognition (NER) and word embeddings, alongside powerful large language models (LLMs) accessed via APIs.

• Machine Learning Predictions: Using node and graph embeddings (via Node2Vec), the app predicts which notes are likely to be connected based on the content.

• Displaying related notes to each note using the graph structure and KG Embeddings to predict edges. Notes predicted to be highly similar are directly connected through a virtual/predicted edge.

• Adding notes and information in a ChatGPT-like manner or like a "saved messages thread" with automatic context linking (Context Lynx = Context Link).

• Supported Inputs:

  • Plain Text: Write or import plain text notes.
  • Web Links: Automatically scrape and extract content from websites to capture information.
  • YouTube Links: Extract transcripts from YouTube videos to integrate their content into your knowledge graph. For the scope of the project, integrating a paid API for reliable transcript access was not included; in the worst case, only the link to the YouTube video will be added as a note.

• Showing all notes ordered by creation date.

Figure 4: List of all created Notes, a Note can be selected to show related notes

Technical Details

• Backend: Written entirely in Python, ContextLynx combines local machine learning models with API-based access to large LLMs for context generation. The application is built on the Django framework.

• Graph Embeddings: Utilizes Node2Vec for generating node/graph embeddings, enabling ML-driven predictions about note relationships.

• Course Project: This project was developed as part of the "Knowledge Graph" course at TU Wien.

Deployment

You can try out ContextLynx live at contextlynx.com/get-started.

Dependencies

• Docker

Local Deployment

To deploy ContextLynx locally, you'll need Docker installed on your machine. Follow these steps:

1. Clone the repository.
2. Copy the environment template file .env.template to .env:

   cp .env.template .env

3. The OpenAI API key is not included in the repository. You'll need to create your own API key by signing up and generating one at: https://platform.openai.com/organization/api-keys
4. Start the service using Docker Compose:

   docker-compose up --build

5. Visit http://localhost:8000/get-started