dialoguemd-archives/covidflow

Flow for covid-19 assessment, navigation and follow-ups.

covidflow's Issues

Properly support refreshing the page.

Currently, when we refresh the page, the enabled/disabled state of the input box is reset while the session ID (thus the conversation state) is preserved.

We should either completely reset the conversation with a new session ID or preserve the input box state.

Send inital request when starting the chat widget.

The chatbot currently expects to begin with the /greet intent. We should also avoid persist the history when the tab is closed and simply use session based storage.

Basic self-assessment workflow

Merge utter actions in tested positive flow stories

Create a FAQ intent detector and integrate it in Rasa NLU pipeline.

This could be done by checking the presence of a question mark, the absence of entity, the number of words, etc...

Change rasasdk Action and QuestionAnsweringProtocol to run async

Current implementation of these are run synchronized which will affect their responsiveness when performing slow operations such as http requests.

For the rasasdk, we could reuse the async objects from beautiful-mind.

Extract E2E test framework + initial happy path test suite

The code will stay internal to Nu Echo and run locally until ready to be put on GitHub.

Flow: Update and clean-up messages

Missing the 211 messages. Clean-up messages: join similar messages together when the meaning is the same and rename messages to standardize naming

Action Server: Send phone validation code

Code will be sent using an external event on Twilio channel

Action Server unit tests can't run in CI

Any test that depends on rasa-sdk will fail in the CI.

This seems caused by the fact that we don't install rasa-sdk. We should probably add it in our dev dependencies instead of installing it in a separate target.

Configure separate Twilio numbers for french and english to properly support Inbound messages.

Flows: separate fever and update other symptoms messages

Maybe make a little form out of asking severe symptoms? Or more? The slots in which to keep the symptoms also changed and the mild symptoms message

Extract name entity in daily CI enroll flow

When asking the user for its first name, we currently set the first name to the text value of the complete user message. We want to extract a first name entity instead.

Edit: We will not implement this for now. We will see if users actually enter something other than their first name, then reassess.

Daily checkin enroll flow

Connect UI frontend with Rasa Core

Deployment: Automatic table creation/migration

• User profile table
• Daily assessment result table
• Validate that we can create other tables chloe database

Disable text input when not asking for text input.

To avoid managing error handling everywhere or providing a bad user experience, we must find a way to disable the text input in our chat widget when the output contains buttons.

Core + Action Server: Send reminder SMS upon reception of an external event

Propose info package in assessment

Asks for province and age after severe symptoms, proposes info package instead of redirection websites

Flow: update daily CI enroll flow to match v0.6 of design

Remove the time of day slot and add the Dialogue account one, plus remove the email option. (only SMS will be supported for now)

Regex entity support

We need to extract some entities that cannot be expressed as buttons (phone number, email address, first name, messenger id). To do so, we should setup a simple regex based entity extractor (instead of using a duckling server).

Core: Support Twilio channel (+ dev account support)

Action Server: Self assessment persistence in DB

Extract the timezone from the browser

Pass it to the Core server as an entity and store it as a slot.

Separate Rasa Core and Action Server in separate projects to ease deployment

Update suspect/get_assessment flow to fit v0.8

Deploy Action Server using CircleCI

We must also connect it to the Rasa Core instances using ACTION_SERVER_ENDPOINT.

Flow: implement assessment after QA

Entering the QA flow after an assessment is possible, but the opposite should be true in some situations, and is not implemented.

Use poetry instead of pip

We still need to figure out the proper workflow with our Docker images that uses base Rasa images. We might want to use rasa and rasa-sdk as dev dependencies to avoid reinstalling them. Else we might want to take the custom Docker image route.

Kubernetes Cron Job: Custom Docker image that triggers the external events

Merge utter actions in suspect flow stories

The language should be part of the URL context path to allow starting the application using the right language from a SMS sent URL.

Tested positive flow

Returning for check-in flow

Frontend: Special route + intent to trigger daily check-in

Pass the user language to the Q&A service.

While the Q&A service has language detection features, we should still explicitly pass the language when querying it. To do so, the user language must be available in the action server request using a slot which means that the core servers must be aware of their language.

To do so, we should add a language slot to the domain file and initialize it with the proper language:

slots:
  language:
    type: unfeaturized
    initial_value: en

The value could either be specified in separate domain.$language.yml or using an environment variable if possible.

Create module to communicate with Q&A service

The service URL should be configurable using an environment variable.

Make language configurable

The frontend language must be configurable like in the current Chloe implementation. We will need to point to the the correct core instance (en/fr).

Ensure the whole application is HA compliant.

Some Rasa features (like reminders which are stored in a specific instance) are not HA compliant. We must ensure that our application scales (both up and down) properly and that we don't have issues when a node goes down.

We might also have issues with Websockets.

nginx.ingress.kubernetes.io/session-cookie-change-on-failure and DB tracker store are steps in the right direction.

Daily check-in demo flow

Implement the first message and first question of the daily check-in flow

Format scraped data properly before sending it to the user.

Data may contain HTML tags for example. We should be able to convert some into the markdown format supported by the Chat Widget.

Publish E2E test framework to PyPi and add a dependency in covidflow project.

What's missing:

• License (Apache 2.0, MIT, other?)
• Accurate documentation
• PyPi publication

Make Rasa core debug mode configurable using an environment variable.

We currently start the Rasa core applications using the --debug flag (in our Dockerfile) which generates a lot of noise that is not parseable by the Datadog log engine.

We should make this configurable using an environment variable (or maybe simply change the endpoint when running them in our docker-compose.yml).

Integrate QA flow

Find proper values for session configuration

We should properly end the conversation after a predetermined duration.

See the documentation for more details.

Deploy Postgres tracker store

Deploy Rasa Core using CircleCI

Action Server: Load user profile from DB to prefill slots when starting daily check-in

Core: Daily self-assessment flow (upon receiving a specific intent)

Deploy Rasa Core (fr) using CircleCI