DeepLearning in Microservices

Implementation of TensorFlowJs models in NestJs microservice architecture

Description

This project consists of two microservices. The first microservice is called App and acts as a gateway. The App receives user's Text and sends it to another microservice called Ai through a gRPC connection. Ai analyzes the text and returns toxicity as a boolean. Also, the type of toxicity is described in six different categories: identity_attack, insult, obscene, severe_toxicity, sexual_explicit, and threat. The results are produced from a pre-trained network in Pre-trained TensorFlow.js models, specifically the Toxicity classifier. You can substitute your own model with the toxicity model and use this API for any custom application.

A straightforward illustration of the project is presented below:

Give it a Star! ⭐

If you liked the project and find it useful, support me by giving it a star 😉

Installation

Clone the project

git clone https://github.com/armin3731/nestjs-tensorflowjs-microservice

Go to the project directory

cd nestjs-tensorflowjs-microservice

Install dependencies

npm install

Environment Variables

To run this project, you will need to add the following environment variables to your .env file (or simply rename .env.example file)

AI_HOST

AI_RPC_PORT

APP_HOST

APP_HTTP_PORT

APP_RPC_PORT

Running the app

First, you need to run the App microservice:

# runs App microservice as default
$ npm run start

# or
$ npm run start app

Then open another terminal and run Ai microservice:

# Ai microservice
$ npm run start ai

Note! Starting the AI microservice will take some seconds because it loads the model at the start.

Note! To my dear friends from Iran. Use a tunneling system while running Ai microservice, because Google Cloud won't let you download the model.

If you use .env.example as your ENV variables, you will see a swagger API documentation on http://localhost:3000/api

API Reference

Analyzing Toxicity

POST /analyze_toxicity

Parameter	Type	Description
`text`	`string`	Required. The Text you want to find out if it's Toxic or Not

Screenshots

ToDo

Caching the model to start faster
Migration to Nest v10

Tech Stack

ML: TensorflowJs

Server: Node, NestJs

License

MIT licensed.