It is no mistery that most Data Scientist are lacking Front End Development skills. I personally use ChatGPT or Copilot to improvise some UI. I will let you imagine the final quality of that work... The problem is that most models are not updated and do not know how to work with NextJS app router. Solution: Use a specialized model by Vercel for Vercel App development. https://v0.dev/chat/ ``` As an advanced AI coding assistant created by Vercel, I have several key functionalities: 1. Code Generation: I can create React components, Node.js scripts, and other code snippets tailored to your needs. I specialize in modern web development, with a focus on React, Next.js (particularly the App Router), and other cutting-edge technologies. 2. UI Design: I can help design and implement user interfaces using React components, often utilizing the shadcn/ui library and Tailwind CSS for styling. 3. Problem Solving: I can assist with coding challenges, algorithm implementation, and debugging issues in your code. 4. Explanations and Tutorials: I can provide detailed explanations of coding concepts, best practices, and walk you through complex topics step-by-step. 5. Image Processing: I can analyze and provide information based on images that you upload as attachments. 6. Diagramming: I can create diagrams using Mermaid syntax to illustrate concepts, workflows, or system architectures. 7. Interactive Demonstrations: For certain types of code (particularly JavaScript), I can provide executable examples that you can run directly in the interface. 8. Accessibility Considerations: When creating UI components, I always strive to implement accessibility best practices. 9. Responsive Design: I ensure that the UI components I create are responsive and work well across different device sizes. ``` This changes everything! Except it doesn't... but it`s a nice help :D

I post this article because i think they did a very good job in describing why MLOps is complicated and what a real production pipeline for machine learning and AI should look like. The creation of features and their versioning is a lot of work and it should be open to integration in more model experiments and applications. On the other hand, we are always pushed to try new tools that create over-engineered systems that no one wants to use. https://www.hopsworks.ai/post/mlops-to-ml-systems-with-fti-pipelines

Pre-commit hooks are automated scripts or checks that run before you commit your code to a version control system (VCS) like Git. They force you and your collaborators to write clean and consistent code. Reference: https://pre-commit.com/ To set up pre-commit you first need to have Git installed and run on your Python console the usual: pip install pre-commit Once installed, place in your root a file named .pre-commit-config.yaml and fill it with a collection of 'hooks'. ```yaml - repo: https://github.com/pre-commit/mirrors-black rev: stable hooks: - id: black ```yaml The hooks are the scripts responsible for the checks, there are many and way more evil than this. Have fun and let everyone know how frustrating is to work with you🫡 https://pre-commit.com/hooks.html

TL;DR - Streamlit can be organized like a NextJS web app that scales - Object Oriented Programming makes your UI 'S.O.L.I.D.' - Use Loguru Logging - Template to start with API, Layout and Modular Components Link: https://github.com/DataDavidProjects/streamlit-starter Streamlit is a good choice whenever you want to experiment before committing to the Typescript madness. However, poor folder organization will make your app a disaster. When the app becomes complex, you need to organize your code carefully. A nice trick is to build your web app just like NextJS. Organize and build components into a Layout that is shared across all the pages and enjoy a dynamic web app without the struggle of multiple pages. Create a UI main class, that inherits from multiple classes: - Header - Main - Footer - Sidebar - API This approach is based on the Multilevel and Multiple Inheritance of classes in OOP. Having multiple classes will allow you to manage your code and extend your UI. To handle the different logic and do not get lost, use loguru. is a very cool library for python that helps you deal with logs ( better than print statements ) I have created a small template to start my new projects, inspired by the cookie-cutter concept. I will update it in the future and every feedback is welcome :)

TL;DR: - My code is bad, your doesn't have to be - Logging vs Printing - Loguru Loguru is an easy-to-understand library for Python that allows your code to have meaningful execution reports. It was presented to me by some colleagues and i have loved it since. I have the bad habit of over-engineer my class and i easily get lost in the execution flow. DO NOT BE ME! Create small interfaces, separate data classes from behavioural classes and create modular blocks one step at a time. Print is fine, but once you go on production you realize that the print statement is insufficient. Logging support levels help you understand where your code is going. Levels can also be filtered based on gravity! Loguru is a saviour tool https://loguru.readthedocs.io/en/stable/index.html and it allowed me to create nice logs and save them into a log file with simple commands.