This is truly mind-blowing to watch a genius in motion – running through of how to build up an AI start-up in 40 minutes.
In this video, he breaks down to 10 steps to build a complete start-up like this.
- Step 1 List Personal Problems
- Step 2 Market Research (competing products)
- Step 3 Buy Domain
- Step 4 Create Landing Page
- Step 5 Share Landing Page
- Step 6 Create Business Plan
- Step 7 Design Pipeline
- Step 8 Transfer Learning (!)
- Step 9 Create Web App
- Step 10 Deploy!
Then at step7, design pipeline:
- User Authentication (SQL)
- Database (SQL)
- Payment (Stripe)
- Uploads (Flask native) (web development for Python, also Jango etc.)
- Inference (Keras)
- Authentication + Database + Payment functionality https://github.com/alectrocute/flasksaas
- Upload functionality https://github.com/bboe/flask-image-uploader
- Inference? Keras
Next, for image diagnosis, the AI technology we gonna use is convolutional neuron network using Keras deep learning model. Google already has done the heavy lifting by training giant datasets, so the best way is to utilized it. Also another great channel is borrow the codes from Kaggle. So Saraj pulled the trove of image data from Kaggle. He also demoed to search “flasksas” in github, His codes are saved in https://github.com/llSourcell/AI_Startup_Prototype/blob/master/Transfer_Learning.ipynb
cd the directory where he stored the codes for the sql database construction, then run commands on terminals directly:
python3 manage.py initdb
python3 manage.py runserver
For the last deployment step, just go to Heroku, Heroku platform uses Git as the primary means for deploying applications (there are other ways to transport your source code to Heroku, including via an API”. In terminal type “git push heroku master” .
Furthermore, one can think of the mobile app, add batch inference, better design, using tensorflow serving (libary) with version control system, continuous training over time to finetune the image diagonsis.