Supporting Development of COVID-19 Assistants

Hello Rasa @community

Like most companies, we are searching for a way to be helpful as the COVID-19 pandemic continues to unfold. In our case, we have been grateful to support the development of AI assistants to help disseminate information about the virus, how to stay safe, and where to seek help.

We’d like to share some COVID-19 resources that may be helpful if you’re either planning to build an AI assistant or use one to stay updated with the latest information:

  • A health alert service from the World Health Organization (WHO): a free messaging service that provides the latest news about the virus and numbers in real time, including information on symptoms and how to stay safe. This service is available in Arabic, English, French, Hindi, Italian, Spanish, and Portugeese.
  • An open-source COVID-19 Chatbot Model from QBox: An AI assistant starter pack that addresses the most common questions about the COVID-19 virus.
  • An open source COVID-19 model in Lithuanian.

If you’re interested in what’s been going on, or have additional resources to share that would benefit the community—or are working on a COVID-19 assistant and want help or inspiration—post them here in this thread. We hope everyone is staying safe and healthy.


It has oly intents and entities ie only nlu data .How can i see the bots working it doesnt have utterences only…

1 Like

Hola from Córdoba, Argentina!

We have an assistant who responds to COVID-19 questions and questions about the measures the goverment is taking with respect to the pandemic. check it out: The (not-for-profit) project is going really well we have a huge team of volunteers, journalists, epidemiologists and computer scientists. I’m leading the backend development part with Rasa.

I have done other task-oriented bots with Rasa that worked really well, but this one is different from a design point of view. Because the stories are all short, one question, one answer, it is much simplier than, say, a scheduling bot. However, the challenge is that we have such a high number of intents, more than 50, that it is getting hard to train for a model that knows the difference between “Are the hardware store open?” - “¿Están abiertas las ferreterías?” and “Are the grocery stores open?” “¿Están abiertos los supermercados?” And more unanswered questions come in everyday that we need to incorporate into the system.

Any advice on how to maximize the NLU performance, even if it means ignoring the entity extraction and/ or taking focus away from rasa core (story) training?

Here’s my current pipeline:

language: es


  • name: SpacyNLP
  • name: SpacyTokenizer
  • name: SpacyFeaturizer
  • name: RegexFeaturizer
  • name: CRFEntityExtractor
  • name: EntitySynonymMapper
  • name: SklearnIntentClassifier intent_tokenization_flag: true intent_split_symbol: + policies:
  • name: MemoizationPolicy
  • name: KerasPolicy
  • name: MappingPolicy
  • name: EmbeddingPolicy
  • name: FormPolicy
  • name: TwoStageFallbackPolicy nlu_threshold: 0.2 core_threshold: 0.2 fallback_core_action_name: action_default_fallback fallback_nlu_action_name: action_default_ask_affirmation

Any feedback would be amazing!


1 Like

I was quite excited to see An open source COVID-19 model in Lithuanian until I clicked the link and realized that it is my own repository that I used to experiment with Rasa and hadn’t touched for 2 weeks.

I’m sorry, but it isn’t a serious project, please remove the link from the first post.

1 Like

I’d recommend using retrieval actions; that’s their intended use case.

I whipped up a quick repo with the data in that format, for your use. Unfortunately had to use GPL license as the original repo had it.


Hi Mady,

Hope you are doing well. I am Palash Nandi from India, we are interested to build this assistant bot in Indian local languages. As you have mentioned that resources are in ‘English’ and ‘Hindi’ which we can understand ( can translate in other Indian languages though ), would love to have that. Please let us know what to do to get those resources.

Be Safe :smiley: .,
My email id is : Thanks in advance.

1 Like

@niveK You are a genius! I’ve wanted to implement chitchat but I never understood exactly why it is used. Now I do! Thank you so much for the repo. We will definately be trying this out.



@bubjanes Hello Brandon, I went through your chatbot and its really impressive. We were also expecting to do a bit same like it as we had same requirements, for COVID-19 awareness. I would be highly obliged if you could share your code with me.

I am stucking in how to create a flow based on user selection of button/option in form.

Hola @pranay_raj!

Thanks for the kind words. We are about to publish our newest version with DIET. That changes everything! I’d be glad to share some of my code. Would you like me to email you? I’d rather not clutter up the forum.



1 Like

Hola again @bubjanes ! Sure, i would love to receive them on email, my email id is . I would be waiting for your reply. Thank you so much for the same.

1 Like

Hola @niveK! I tried a version of my COVID-19 bot with retrieval actions and I came across a problem: apparently the retrieval actions does its classification the training questions in the nlu data (as is expected) but also using the response, which is a weird way of designing it. I kind of wish that it didn’t do that. As a result, I think, the intent classification failed more often with retrieval actions than if we took the DIET intent classifier by itself. I used triggers from the Mapping Policy and had better results.

Any thoughts?

I can’t speak to the design of the training, but I don’t think I saw issues with triggering the respond intent. The performance impact could have been as a result of having omitted the stories, and would have been resolved by adding Mapping Policy triggers for the respond_faq or respond_company actions.

So after adding the retrieval actions training data, you would have needed to add something like:

## respond to FAQ
* faq
    - respond_faq
## respond to company inquiries
* company
    - respond_company

Regardless, glad to hear it’s working!

1 Like

@bubjanes Hey Brandon. Were you able to mail???

1 Like

Hey @pranay_raj!

Sending now!

Hi, I’ve created an api for finding medical locations where you can get tested for COVID-19 and created an example bot with Rasa.

Feel free to grab all the resources and take it to the next level :muscle:

I would like to work on this project further, add more locations and a more fine grained search. I’d be more than happy to team up with either a developer or a data provider.

Thank you and keep safe :slight_smile:

1 Like

Super interesting, Tomas. Thanks for sharing. Let me know if you’d be ok with us sharing this on our landing page. Thanks!

Thanks for sharing Brandon! We’d love to include this on our landing page as well. Let me know if you’re ok with that.

Thanks for reaching out Palash. The WHO health service is available in Hindi and other languages. We don’t have access to the source code yet. We’ll publish it as soon as we do.

That would be great! Aguante Rasa!

1 Like

Hi Mady, sure thing, I’d feel honoured. Glad you find it interesting.

1 Like