System to build, manage and deploy classifiers.

• Join Slack
• Trello feature board
• Trello development board

• User authentication through registration, or OAuth2 through google & github

• User interface to allow non-technical users to build categories of data and train machine learning classifiers

• Provide tools to support data collection

  • Integrated Jupyter notebook hosting
  • Image catalogue
  • Google image parser
  • Image manipulation
  • visualization

• Allow technical users to upload their own classifiers of any type

• API access to classifiers through owner defined JSON interfaces

• Versioning & rollback capability on classifiers & APIs

• Micro-service architecture approach to hosting these APIs - each endpoint should live in its own "pod". It should be able to be updated, restarted and scale independently without affecting any other part of the system

• APIs have the capability to accept user feedback and retrain themselves on either feedback thresholds, time intervals, or on the owner's instruction

• APIs keep metrics on their usage, performance and behavior over time

• more TBD

1. Pre-install requirements:

   • Docker
   • Node 8+

2. Start API server:

Fill out the .env.template file with variables and then copy it as .env in the root directory - this file is git ignored.

Then in the api directory:

docker-compose up

browse to http://localhost:8000/api/v1/

3. Start node server:

Starting in the root directory in a new terminal:

cd frontend
npm i
npm run dev

and browse to http://localhost:3000

Warning: This environment is not ready for production (hard coded variables, secrets, etc.). Only used for development purposes right now

Primary prerequisites:

1. Minikube
2. Kubectl
3. Telepresence

minikube start

# open the kubernetes dashboard in your browser
minikube dashboard

# build starting API & web docker containers
bash build.sh

# Launch all services into the minikube Kubernetes cluster
kubectl apply -R -f k8s

# retrieve the URL of the API gateway
minikube service ambassador --url  # open this in your browser

All services in the Kubernetes environment are behind an API gateway, Ambassador. Instead of making a code change, rebuilding docker containers, and then applying changes to the K8S cluster, we can write our code locally and have it apply inside the K8S pod it is running in. This will simulate an exact production environment.

To do this we use Telepresence. Below is an example command to proxy our local api service into the K8S service. From the root directory:

telepresence --swap-deployment mako-api --expose 8000  --docker-run  -it -p 8000:8000 -v $(pwd)/api:/code --entrypoint "/bin/bash" mako_api:latest
# or run this in the bash script in the repo
bash dev.sh

Join & ask in slack! General guidelines:

1. All work is subject to reviews before merge
2. Work in feature branches
3. Rebase your branch on top of master, we will try to maintain a clean git history
4. Utilize linters & styling tools. TSlint & Prettier for the frontend, PEP8 for Python

• Backend API: Django, DRF, Postgres, Redis, Celery
• Frontend: React, Redux, TypeScript
• ML: Pytorch