I have been working in AI research for over a decade now. Most of my work was conducted in academic and research lab settings. I love thinking of new ideas and prototyping them to verify that they work. However, lately the research landscape has become even less friendly for researchers not working in “Big Tech”. In my quest to (finally) make a move to the super-fast-moving industry, I realized that every company is interested in an end-to-end development pipeline. It’s one thing to read Machine Learning System Design books, but I decided the best approach was to build a complete, end-to-end AI product in public, focusing on the MLOps and production-readiness skills.
Perhaps I can make another post about my interview experiences so far and how ridiculous and archaic are the standard hiring practices.
However, this post, and hopefully the following posts in this series, are focused on a pet project that I decided to implement. I will document how I am going about this project as it progresses. So, what is this project and why did I choose it? A few days before Prime Day 2025, I decided that I wanted to buy a robot vacuum with mop. I started my research by scouring the internet to find which machine to buy. Searching through Amazon product pages and reviews, Vacuum Wars, RobotVacuums subreddit, and a plethora of Youtube reviews, I ended up being far more confused than at the beginning. Even the best of deep research LLMs were not of much help. I really don’t want to read a 10 page article and end up with “it depends on your needs”. Of course it does! So, I decided that I will create a system whose only job will be to recommend the right robot vacuum for your needs.
I am going to document my journey to creating this system, and hopefully it will be helpful for some. Before diving in, I want to say a few things about this process. These posts will not be AI written. I want to convey what I actually learned, and writing about my project will make me rethink about some of the choices I made. I will also not be asking an LLM to make this project for me. My experience is that LLM’s don’t really do very well when it comes to large projects, and also that is not the goal of this exercise. I will, however, use web search and LLMs to learn about new technologies that I may be unfamiliar with, and learn about best practices.
For this project, I plan to use a modern, serverless stack on GCP and emphasize best practices for MLOps and automation. Based on my initial thoughts, I will be using the following key technologies:
- Python: The foundation for all data processing and model development.
- Google Cloud Platform (GCP): For the entire infrastructure, from serverless data ingestion (Cloud Functions) and warehousing (BigQuery) to model training and deployment (Vertex AI).
- Web Scraping
- Natural Language Processing (NLP)
- Docker: To containerize and serve the final web application
- GitHub Actions: To build a full CI/CD pipeline, automating testing and deployment from day one.
- Streamlit: My current front-runner for quickly building an interactive front-end.
I think this project can be of interest to people who are curious about AI development, cloud implementations, robot-vacuums, and technology in general.
You can also follow the code development on GitHub: https://github.com/asadanwar100/robot-vacuum-ranker
Top comments (0)