My first AI product - Interiobot.com
Data Eng Weekly - Ep 13
Surprise !
I am happy to share some exciting news with you today - I have just launched my new AI product, interiobot.com!
This tool is designed to automatically generate interior design concepts based on a photo of a room.
It's perfect for anyone looking to bring their interior design ideas to life quickly and easily.
interiobot.com is also aimed at real estate professionals who want to improve the conversion rate of their property listings.
You can try for free : 6 designs are on me :)
Examples of design generated (on the left the original image, on the right, the generated one)
Behind the scene
As I was building interiobot.com, I realized that I was learning so much along the way - things that I believe could be valuable to others in the same field.
So, I wanted to take a moment to share some of my key learnings with all of you.
It took me approximately 5 weeks to build the prototype, with an average of 15 hours of work per week (~2 weeks full-time).
While I was able to handle most of the development work on my own, I needed a little extra help for the front end. I got the help of a freelancer (~35h) to build the React app, as I didn't have a lot of experience with that framework.
I started to use GTP-4 this week to help me with the front end and I have to say I was impressed with the results! It takes the same amount of time to define requirements to a freelancer and the output is comparable except that the iteration speed is crazy fast.
You want to change a color, a size, a distance? just “prompt it” and you have the result: no more back and forth with another dev.
I tried as well to use it for infrastructure code generation, and the results were really good.
No more time spent on searching bad AWS documentation: get a first draft by chat-GPT and fine-tune it.
Try by yourself: “serverless.yml code of a lambda triggered by a new object in an s3 bucket”
For my next product, I will probably not use any external help, but instead, use chat-GPT more intensively as my “intern”.
Tech Stack
The goal I had when building interiobot.com was to make it 100% serverless. Why ?
I wanted to keep costs down as much as possible. By building a serverless app, I knew that I could avoid paying for idle server time and only pay for what I used.
it makes it much easier to scale the app up as needed. This means that I can focus more on building and improving the app itself, rather than worrying about database management for exemple.
Here is the complete breakdown of my stack:
React for the front-end + Tailwind CSS (which reconciled me with CSS btw :) )
AWS Amplify for the app setup
AWS Cognito for user pool management
AWS AppSync (GraphQL) for the API layer
DynamoDB
Step Functions for backend workflows: image processing and payments management
Serverless Framework for infrastructure as code
Replicate for GPU serverless inference
Serverless GPU inference
I underestimated this part of the app. AWS does not provide an out-of-the-box serverless GPU processing service.
To solve this problem, I tried out three different external providers - beam.cloud, banana.dev, and replicate.com. After testing each one, I ultimately decided to go with replicate.com.
Beam.cloud had some limitations that didn't work for my needs, such as not offering the A100 GPU and having a long cold-start time. Banana.dev, on the other hand, wasn't as developer-friendly as I had hoped (dependencies were difficult to set up in my case).
Replicate.com, however, was a great fit for my needs. They offered the A100 GPU and had a fast cold-start time, which was essential for the performance of interiobot.com. Plus, their platform was easy to use and integrate into my existing setup.
I'll be continuing to share updates on the development process of interiobot.com over the next few weeks. I'll be sharing more about the feedback I'm receiving, any updates or changes to the app, and any other interesting tech insights that come up along the way.
I have plenty of ideas to make the product evolves:
furniture visual search with Google Lens
editing of the design using inpainting
and much more…
I'm really excited to keep pushing forward with this project, and I can't wait to see how it evolves over time :) So, be sure to stay tuned for more updates - I'll be keeping you all in the loop!
And last but not least, feel free to try the tool and share it with the people you think might be interested.
I'm always looking for ways to improve and make the app better, so your feedback is incredibly valuable to me.
thank you for reading.
-Ju
I would be grateful if you could help me to improve this newsletter. Don’t hesitate to share with me what you liked/disliked and the topic you would like to be tackled.
P.S. you can reply to this email; it will get to me.