Model Master

A comprehensive collection of AI models, catering specifically to AI engineers seeking models based on their requirements in various fields.

Comment 1

For Judges

Remember to read the instructions on our Github before accessing our website, thanks!

New features for finalist improvements are listed at the end of our project description!

Three important notes to allow best experience:

  1. Run RESTapi.py following instructions in Usage section in Github README.
  2. Website supports resolution of 1440px or more.
  3. To use Favorite Model function, please login with our authorized account:
email: ali.daixin.tian@gmail.com
password: 12345678

Inspiration

As a researcher and an engineer in the machine learning area, many of us often face challenges finding a perfect model for the project. We also sometimes find it hard to gather relevant ai model information and do comparisons in research. Our development team find a gap in collection of models and efficient management and search platforms in current market.

In a rapidly evolving world driven by technology, the potential of artificial intelligence knows no bounds. Imagine a vast repository of AI models, a virtual treasure trove, waiting to be explored. A place where you, the AI engineers, can find the perfect match for your projects effortlessly. This is what we aspire to create - a comprehensive website that serves as a hub for all things AI models.

Our ambitious project aims to empower AI engineers to shape the future in a technology-driven world. We envision a comprehensive website, serving as a hub for a vast collection of AI models, where engineers can effortlessly find the perfect match for their projects. With our platform, we strive to equip AI engineers with the tools they need to push the boundaries of innovation, be it in computer vision, natural language processing, robotics, or other cutting-edge fields.

What it does

  1. Extensive AI Model Repository: The website hosts an extensive database of AI models from diverse fields, such as computer vision, natural language processing, robotics, and more. AI engineers can explore and find models suitable for their specific needs.

  2. Targeted User Base: Our platform is specifically tailored for AI engineers who seek efficient ways to discover AI models aligned with their project requirements.

  3. TiDB Serverless Database: We utilize the TiDB serverless database for its scalability, high availability, and cost-effectiveness. This ensures smooth and reliable management of the vast collection of AI models on our website.

  4. Intelligent Search Functionality: The website incorporates an intelligent search function that understands user input prompts, thanks to TiDB’s chat2query. AI engineers can describe their desired model, and the platform will provide relevant suggestions and matches.

  5. User-Friendly Interface: We prioritize a user-friendly interface, ensuring that AI engineers can navigate the platform effortlessly, find models quickly, and access essential details for each AI model.

How we built it

1.We utilized the BeautifulSoup and requests Python framework to scrape data from the Hugging Face, obtaining information including name, model size, popularity, link, task, languages etc.

2.We employed the serverless database, our TiDB to store all the information. To implement all the information to the front end, we used the Python framework Flask to construct a REST API. This API acts as a local host, receives HTTP requests, processes them through our custom API, gains access to the TiDB C2Q endpoint, and ultimately outputs an HTTP response message.

3.With the http response from our api, we are allowed to use Webflow content management system and javascript to show all contents in the webpage.

Challenges we ran into

Lack of experience in database development It is our first time doing project with serverless database, we have zero knowledge on SQL or database development. This is a big challenge for us but also a great opportunity. Thanks to TiDB’s chat2query we can get a basic view of sql and how to implement that to our project.

Time Management To be honest, we start our project 10 days before the deadline which is 6 times shorter than the given time 2 months. But our team puts great effort and spends on average 6 hours per day per person before the deadline to make it possible!

Web Design

  1. Time-comsuming graphic processing on svg icons;
  2. Large CMS data which requires extra time on importing and pre-processing

Accomplishments that we're proud of

1.Our proudest achievement is the successful creation of an expansive AI model database and a user-friendly, efficient platform for AI engineers. Through our website, AI engineers can effortlessly find suitable AI models tailored to their project needs.

2.Furthermore, we optimized our search functionality to comprehend user input prompts, thereby providing relevant suggestions and matches.

3.We have made a scraper for model collections, along with a large-scale dataset from Huggingface.

What we learned

In this project, we learned how to leverage BeautifulSoup and requests Python framework for web scraping to acquire relevant AI model information. We also learned how to use the serverless database, TiDB, for massive data management, how to use built-in chat function Chat2Query and how to build a REST API using Flask, facilitating effective transmission from the back end to the front end. Additionally, we learned how to design and implement a user-friendly and efficient website interface, using Webflow CMS and javascript to enable AI engineers to easily find the AI models they need. We deeply understood that every step, whether handling large data or creating a user-friendly interface, requires careful design and consideration.

What's next for Model Master

  1. Build dummy server: As an initial step, we will construct a dummy server to serve as a testing and development platform. In the future, we plan to deploy our application to an actual server. This iterative approach allows us to iron out any issues and ensure the system's reliability before launching it on a live server.

  2. Dynamic examples: We will enhance our homepage by improving 12 examples of AI models. These featured models will be dynamic, meaning they can be updated or changed according to our strategic decisions, such as highlighting trending models, newly added models, or models from diverse categories. This will provide a snapshot of our expansive AI model collection to users visiting the homepage.

  3. Favorite Models: To improve the user experience, we plan to add a "Favorite Model" function. This will allow users to save models they are interested in for easy access in the future. By marking a model as a favorite, users can quickly retrieve and review it later, enhancing the overall navigation and efficiency of our platform.

V2.0 - 2023-08-06 - Hackathon Finalist Improvements

New Features

  1. Login/signup function connected to TiDB database
  2. User customized favorite model list function

How we build it

We use the power of TiDB to drive our account management and user favorite functionality. We build a user_info table in our database to store all user information along with their favorite model list, where we can also execute ADD and UPDATE and other SQL statements for our different functionalities. TiDB's remarkable scalability and real-time capabilities have enabled us to seamlessly handle the complex task of managing user accounts and facilitating their personalized experience through the favorite feature.

Optimizations

  1. Optimize keyword search
  2. Improve model card rendering
  3. Revise and optimize model abstracts using BART-Large-CNN
Share this project:

Updates

posted an update —

V2.0 - 2023-08-06 - Hackathon Finalist Improvements

  • New Features Login/signup function connected to TiDB database User customized favorite model list function

  • How we build it We use the power of TiDB to drive our account management and user favorite functionality. We build a user_info table in our database to store all user information along with their favorite model list, where we can also execute ADD and UPDATE and other SQL statements for our different functionalities. TiDB's remarkable scalability and real-time capabilities have enabled us to seamlessly handle the complex task of managing user accounts and facilitating their personalized experience through the favorite feature.

  • Optimizations Optimize keyword search Improve model card rendering Revise and optimize model abstracts using BART-Large-CNN

Log in or sign up for Devpost to join the conversation.