If you read my previous post about getting started on Heroku, you might be wondering how to move on to the next step. While I cannot recommend the best next step for your particular use case, I can suggest something which is a good next step from the viewpoint of learning. The periodic table chatbot A little while back I blogged about the periodic table chatbot. I had used the example to show how to store values in contexts. I had also mentioned the following: As you can see, the bot does not provide the actual values. To do that, we need to connect it to a webhook. I want to keep this example simple, so for now we will […]
I created a video to show how you can use the prebuilt smalltalk agent to make your API.AI chatbot smarter in just a few minutes. The basic steps are: Import the prebuilt smalltalk agent into a new agent Bulk copy intents from the newly created agent into the agent you wish to improve That’s pretty much it!
Recently, Dialogflow revamped their documentation quite significantly. In the process, there has been a clear move towards favoring the Google Cloud over Heroku. This is understandable, considering Google acquired Dialogflow recently. 😉 In the process though, the Dialogflow developer community is now required to go through a much more challenging initiation to webhooks, and I had this view confirmed by a few private online conversations too. Earlier, the webhook sample had a one-click deploy to the Heroku cloud, and you could get the example up and running in practically no time. So, I have decided to retrace the old steps, combine it with the online resources that I could find, and tried to somewhat simplify the process for folks trying […]
Did you know that the API.AI context can be used like a dictionary object which can hold a set of key-value pairs? You can use this feature to refer to the value of an entity which was used in previous intents. Let us take a look at an example. Periodic Table Bot The periodic Table Chatbot does something really simple. The user can ask for the atomic number, or the chemical symbol, or the color of any element in the periodic table. In addition, they can use the context (that is, remember the element they talked about in the previous message) and enquire about the other properties. For example, here is a screenshot: As you can see, the bot does […]
How well do you know your API.AI contexts? Watch this video and find out! (I think you might find a surprising thing or two) Sorry about the poor audio. I am still learning how to do this. 🙁 Learn about – What happens when multiple input contexts are set for a single intent? – What happens when multiple output contexts are set by a single intent? – The hierarchy of intent selection when there are multiple candidates based on the active context(s) Also check out this article on candidate intents.
I have deleted obsolete and off topic posts from my website. If you are still interested, you can access this article here.
Occasionally, you come across API.AI features that surprise you (in a pleasant way). Yesterday I found one feature like that. The problem In API.AI, you know that each intent can be hooked up with a webhook call. As soon as the intent is invoked, it will first send all (relevant) information to the webhook URL. The response from the webhook is parsed to output a message to the user. Now, it is fairly straightforward to send the parameters that you collect from the user to the webhook. There are occasions where you generate session specific parameters from the browser’s localStorage, and on first glance it doesn’t look like there is any way to send those to the webhook. Invoking intents by calling […]
A few weeks back, API.AI introduced follow up intents. It was a much needed feature, and I would say it was received with some fanfare. At that time, when I added a section about follow up intents to my step by step guide, I mentioned that you cannot create loops using follow up intents. Recently, I answered a related question on the API.AI forum, and I think this would be a good time to discuss when NOT to use follow up intents. Flowcharts I have spent a little time thinking about this problem now, and at the moment I think there isn’t really a good, well thought out representation for chatbot flowcharts. Here is one which I think is actually pretty […]
You can find plenty of questions about team accounts in API.AI in their discussion forum. I like to add an additional team member team to my account to work on the same agent. Is there a way to create or allow contributors to a project? A small variant of basically the same question: I want someone else on my team to enter data to train up the bot. But I can’t see a basic “share” feature anywhere. Is this locked so only public agents can be edited by more than one person? Till last week, this was a pretty major constraint for people trying to use API.AI for building professional/commerical chatbots. Having helped people as a coach/consultant for API.AI, it […]
I took all the prebuilt agents API.AI made available recently (via their workspace), and created a sort of an inverted index. That is, I created a custom post type called entity where each entity gets its own page. What this means is that if you wish to see an example of an entity you have in mind, or you wish to use an example and see the corresponding entity, you can simply use the search functionality to find it. On top of that, I also went ahead and created a little search engine for those entities (it is possible to customize WordPress search to do this).