Introduction
With the rising interest in Artificial Intelligence and the explosion of large language models, it's increasingly difficult to stay updated on essential research. Eureka Papers is a collaborative project between myself, Tarek Aloui, and Eduardo Pareja Lema, aimed at solving this problem. We are creating a tool to track, recommend, and summarize key AI papers, designed to keep researchers and enthusiasts alike up-to-date.
Accessing Eureka Papers Demo
For those interested in exploring our tool, you can check out the most up-to-date deployed version by following the link attached to this article. However, please note that we are currently utilizing the free tiers on Azure and Vercel for hosting, which means the website may occasionally experience timeouts. This is due to resource limitations on these free tiers. We are in the process of transitioning to a more production-ready environment to alleviate this issue. If you encounter any loading problems, refreshing the website a couple of times should resolve it. Thank you for your understanding and patience as we continue to improve Eureka Papers.
Code
- Eureka Frontend (Nextjs, TypeScript, TailwindCSS, Apollo GraphQL)
- Eureka Backend(Django, Python, GraphQL, PostgreSQL DB)
Objectives
- For Researchers: Eureka Papers helps you quickly identify the most cited and influential papers in the AI domain.
- For Enthusiasts: The tool offers easy-to-understand summaries, making AI research more accessible to newcomers.
Features
- Tracking Most Cited Papers: Eureka Papers identifies the most cited papers in AI.
- Reading History and Recommendations: The tool logs your reading history to suggest new research papers based on your interests.
- Paper Summaries: Summaries for each cited paper are available for quick insights.
- AI Trends Overview: A unique feature that provides a snapshot of current trends and news in the AI field.
Technology Stack
Frontend
Next.js with TailwindCSS and Apollo GraphQL Client (Deployed on Vercel): We chose Next.js for its capabilities in server-side rendering and static site generation, which are crucial for enhanced speed and SEO performance. Apollo GraphQL Client is used to connect to our custom-built GraphQL API, facilitating efficient data retrieval and state management. Vercel was selected for frontend deployment because of its seamless integration with Next.js. TailwindCSS complements this stack by enabling rapid design prototyping without compromising on performance.
Backend
Django with GraphQL (Deployed on Azure): Django is renowned for offering robust and scalable backend solutions. We deployed our backend services on Azure, capitalizing on its cloud capabilities and scalability options. We chose GraphQL over RESTful APIs due to its flexibility in data retrieval, strong type checking, and overall efficiency in handling various requests.
Database
PostgreSQL (Deployed on Azure): Known for its robustness in managing concurrent read-write operations and its extensibility, PostgreSQL was an easy choice for our database needs. We also deployed this on Azure to maintain consistency in our deployment environment.
Additional Technologies
TypeScript and GraphQL-Codegen: TypeScript brings static type-checking to JavaScript, complementing the strong type system of GraphQL. With GraphQL-Codegen, we were able to generate TypeScript types directly from our GraphQL schema, making our code not only type-safe but also more maintainable.
Development Process
Project Management with Notion
Effective project management gains significance when the team size is small. Comprising only myself and Eduardo Pareja Lema, our team has the advantage of simpler, more direct communication. Nevertheless, we believe in the benefits of formal project management methodologies, even for a project of this scale. To this end, we've adopted Notion as our project management tool and aim to align closely with Agile methodologies.
It's important to note that, as we are still students and this is a side project, we are not strictly adhering to Agile principles. However, we are making a concerted effort to stay as aligned with Agile as possible within our constraints. This flexible yet disciplined approach helps us keep an organized record of various tasks, ensuring transparent progress and alignment with our project goals. Though Eureka Papers is still under development, our current trajectory suggests that our project management methods are enhancing both the efficiency and success rate of the project. Additionally, this experience serves as valuable practice in honing our project management skills. We will be elaborating more on our project management experience in an upcoming blog post.
Experimentation with Generative AI
In our project management setup using Notion, we've incorporated an additional layer of efficiency by utilizing Notion AI. This tool has significantly amplified our learning experience. Whenever we are diving into a new concept or technology, we make use of Notion's Web Clipper extension on Chrome to bookmark the relevant tutorials and materials. Subsequently, Notion AI is leveraged to generate a high-level overview of these complex concepts, serving as an introductory guide before delving into the material in-depth.
This unique approach helps us quickly grasp the essence of new technologies or ideas, and it allows for a more effective information digestion. Importantly, this AI-generated summary then serves as a foundation upon which we build our own notes and understanding. We can also hyperlink specific Notion pages that elaborate on the concepts needed for new tasks, serving as a dynamic and continuously updated internal resource.
As we proceed with the project, expect more detailed accounts of how we're utilizing Notion AI to optimize our learning process and development cycle in upcoming blog posts.
Through the creative use of generative AI tools like ChatGPT and GitHub Copilot, along with the thoughtful application of Notion AI for quick and effective learning, we are continually refining our development process for Eureka Papers. This approach has not only optimized our workflow but also helped us improve our prompt engineering skills. These practices contribute to our objective of evolving the project into an indispensable resource for anyone eager to stay updated with the rapidly advancing field of AI research.
Ongoing Work
- Recommendation Model Integration: We are in the process of connecting our backend with a separately built recommendation model that incorporates various parameters to personalize suggestions.
- Automated Database Updates: A cron job is under development to regularly pull the latest research papers from Arxiv, ensuring that our database stays current.
- Social Media and RSS Feed Analysis: We're looking to leverage data from social media platforms like Twitter and Reddit, as well as RSS feeds, to gauge the "trendiness" and potential influence of research papers. This is especially relevant for papers published on Arxiv but not yet presented at reputable conferences. By combining this data with content-based analysis techniques like author and conference h-index and keyword importance measures like tf-idf, we aim to provide a more comprehensive understanding of a paper's impact.
- User Personalization: We plan to implement Next.js Auth for user authentication. Additionally, a large language model assistant is in the pipeline to tailor explanations according to the user's background.
Conclusion
Eureka Papers aims to make the expansive field of AI research more navigable. Through this project, Eduardo and I are learning invaluable skills in both technology and teamwork. We're excited about the potential of Eureka Papers to contribute significantly to the AI community by simplifying access to important research.
By using a carefully chosen tech stack and focusing on user experience, Eureka Papers promises to be a valuable asset for anyone keen to stay updated in the rapidly evolving world of AI.