Building a JavaScript powered Chatbot with NativeChat

Enterprises are increasingly using chatbots to create self-service support options for customers and employees. Chatbot projects are hard to scope out and deliver these days because of the patchwork quilt of technologies required and those technologies’ early state. Most Enterprises do not have chatbot development expertise in-house. Finding such expertise is not easy either since chatbots are early in the technology adoption curve. NativeChat is a chatbot platform providing a declarative process to design and configure conversational flow. I tried out the product by building a NativeChat bot, according to the Car Rental tutorial. It was pretty straightforward and impressive how much functionality I could get in such a short period with no NativeChat experience and minimal chatbot experience. You’ll need to set up a new NativeChat account (it can be a trial account), but it was fast and painless. Important Note: At the time of writing this, the tutorial was slightly out of sync with the UI and the default bot configuration. If you’re trying to do some of the initial tasks and struggling (especially steps that aren’t about car rentals), you can probably skip them. Just try to make sure you understand the basic ideas so you can use them later on.

Highlights

Architecturally, the structure of a NativeChat bot’s “Cognitive Flow”— conversations, reactions, and understanding of relevant data, resembles a declarative finite state machine  like XState or AWS’s Step Functions. Natural Language Processing and the chatbot training you add to the program act as the “events” or “triggers” used by the Cognitive Flow “state machine”. Examples of these are:

• determining what conversation the user is in
• understanding what information the user has already provided
• deciding whether the conversation with the user is complete

Coolest feature corner

NativeChat is great at getting as much information as possible all at once. For example, going through the tutorial, I defined things like how to start a conversation to rent a car and specific questions for rental start dates, end dates, countries, and cities. But just from those definitions, it was smart enough to respond to a user saying something like “I want to rent a car from today to tomorrow in Paris, France,” and it would grab all of the information and jump ahead to asking what kind of car I would like to rent.

General features

• NativeChat comes with a comprehensive and intuitive core set of actions and reactions of a chatbot. Things like conversations, goals, messages, questions, acknowledgments, ambiguities, and suggestions are available to you to use or customize for your configured conversations.
• NativeChat also contains a good set of in-chat UI components called “reply templates” to make it easy for users to know how to reply. These reply templates even offer the user formatted data that might be difficult for the user to input correctly. Dates, times, and other structured formats come to mind.
• Machine Learning or “bot memory” to be used when providing suggestions based on a user’s previous conversational choices.
• The NativeChat product contains robust templating and logical language for response generation, data validation, and data transformation.

NativeChat Console Features

Cognitive Flow

The Cognitive Flow provides an excellent outline view of your current bot definition. Visualizing the conversation structure simplifies building a coherent and realistic conversation.  

Code and Definition Editor

The code/definition editor has helpful features like auto-completion and templates to streamline the work needed to scaffold your conversations, validations, and other elements.

Test Chat App

The “debug” pane gives a ton of useful and easy to grok information. NativeChat developers will appreciate a quick and straightforward debugging process. The “understanding” pane is also helpful 

Debug Pane

Understanding Pane

Training

The training section allows you to import and export your training data into JSON files.

A few features or tutorials I’d like to see

As far as I can tell, these features don’t exist (yet!) in NativeChat. Based on my experiences, I think they’d be useful for more extensive enterprise solutions.

“Pre-validation”/filtering of options

NativeChat supports some “conversationally-specific” validation and feedback like “You can’t rent this particular car in the city you’ve chosen because it’s unavailable in this city.”. However, it would be even better if a user could know in advance that the car is unavailable via the available UI components (or not see it as an option at all).

Split presentation from logic in Cognitive Flow

A lot of the human-readable response text lives in the JSON definition. It would be great if a designer or non-technical business user could easily modify that text without having to redeploy/update the chatbot completely. Large organizations with an international footprint have localization and internationalization concerns. I believe it would be more appropriate for the use case to write a single JSON definition and “fill in” the human-readable parts based on the target locale.

A cognitive flow UI for non-developers

Internally, the organizational processes that would benefit most from a chatbot, are owned by non-technical employees. These employees are the subject matter experts of the process and should be enabled to create, train, and maintain chatbot conversations without going through an Information Technology process. NativeChat goes a long way to simplify chatbots, and if there were a “Wizard-style” way of building out NativeChat bots, NativeChat would be more useful for Enterprises. Similarly, the “outline view” for the chatbot’s “cognitive flow” is excellent. Still, it would be handy to have a UML or similar-style “bot visualization” to make it easier for technical staff to reason about, similar to other finite state machine/workflow engine visualizers like XState’s visualizer. 

A training code editor UI for developers

It’s great that you can import/export training data like “conversation triggers” and “entities.” It would be even greater(?) if the training side of a NativeChat had a UI option similar to the cognitive flow side.

A REST API

If NativeChat has a REST API for cognitive flow updates, training data updates, data syncing, and so on, I haven’t found any documentation about it. Enterprise clients would often prefer to wire the chatbot into their infrastructure via REST APIs.

Bonus – Learn from my example

I’m making my final cognitive flow and training definitions available to examine and use for your purposes. I may have skipped a bit of homework, so you’ll see a fake placeholder email, but it’s close to the final product. You can find them on my GitHub page.