NICOLIS

*Work in progress

You probably noticed the little purple icon in the bottom right corner of your screen, before opening this post.

If you didn’t, now you know that it’s there!

Before reading any further, I urge you to check it out if you haven’t already done so.


I’m assuming you’re not cheating and checked the box before returning.

raunakhede.com is a technical blog. As such, there had been several times in the past, when visitors (especially my mentors) advised me to add an FAQ section. Thus, came the Tech-Queries page (around 2 years back) and a form below it, for users to send in special requests for information that may not have been provided in the post of the project in question.

However, it did not have spam firewalls, as a result of which there were quite a few automated services from websites that bombarded the form service with useless and nonsensical remarks about their own products. After a year of deleting no less than 60 spam messages per week and countering innumerable phishing attempts on my server, the form was removed.

It wasn’t until this month, that the inception of a fully fledged 24×7 Technical Support system to answer frequently asked questions, was put in place. It was initially, built with Dexter. However, due to limitations in scalability, I shifted to Google’s DialogueFlow (formerly API.AI). After a couple of days, I realised that hyperlinks couldn’t be sent in messages to users, which was quite pointless given that cross-referencing to other posts was necessary while framing answers.

The next three days saw a trial of chatbot services and AI assistants like recast.ai, conversation.one, Drift, Social Intents, Chatlio, and even Slack‘s custom integration for a bot on a channel!

Finally, I decided to check out WordPress’s plugins, although the chances of finding one for my specific needs were pretty slim.

It therefore, came as no mean surprise when I discovered the IBM Watson Assistant plugin, which can actually do that and a lot more! After learning the nuances of using IBM Watson on Cognitive Class, I began to build NICOLIS.

NICOLIS is a chatbot with IBM Watson’s machine learning capability, dressed in semi-formal attire.

As is the case with most chatbots, there were several non-technical aspects like smalltalk which needed to be refined to a very high degree.

Fun fact: Many humans can talk about absolutely nothing in particular for hours on end.

But an AI assistant does not function the same way. If you imagine the algorithm to be a waiter in a restaurant, your meal can only be a combination of the things offered on the menu. Out-of-season specialities are rare and a custom-made dish, even rarer.

NICOLIS’s logical and social capabilities are more like that of a jolly waiter, rather than Socrates or Homer.

Nevertheless, NICOLIS has one major badge to wear proudly: a strict ethical code.

Several popular AI’s like Cleverbot (which scored 59.3% in a Turing test against the human score of 63.3%), learn directly from their interactions and consequently, end up repeating phrases or frequently used words in an attempt at social mirroring. This includes unwarranted instances of what would appear to be human emotional outbursts.

Evidently, this is not good for an FAQ assistant’s public relations 🙂

Although NICOLIS has limitations in conversing over local banter,  the weather forecast, modern fashion statements, and philosophy, it is still a cheerful AI with a mild sense of humour.

If you’re a developer/coder/hacker/tinkerer/curious person, go ahead and search for Easter Eggs in the program. I promise, there are no less than five waiting to be discovered!

Here’s how you can build a chatbot with IBM Watson in 18 minutes flat!

*Work on NICOLIS is still in progress. If you run into any glitches, do let me know! I’ll make sure they’re fixed in due course of time.

Leave a Reply