Orion - Portfolio Bot

The Challenge

Create a helpful AI chatbot to live on this very website to act as a guide around my portfolio and professional experience. Must be a little goofy because reasons.

The Approach

In order to showcase newly certified skills in AI Training and Conversation Design, I spent a week experimenting with the Voiceflow platform and building the bot from a few rudimentary commands and slowly expanding on how it responds to input. Some are carefully guided responses; some are happy accidents due to the NLU model being better than expected. The build approach allows for intents to be isolated, make their management fairly easy.

The Final Product

Orion is a little quirky, not perfect, and very much a work in progress. He definitely needs more listening steps and freedom to make decisions based on free input. For a draft experience, the little guy isn’t bad. I am actively having design friends talk to it so I can learn how it responds to unexpected utterances.

What’s Next

This was great for me to cut my teeth on building a working bot. Planned enhancements include having Orion ask for and remember user’s names across multiple visits and have him handle scheduling meetings himself.

Meet Orion

I developed and designed Orion with a sci-fi theme, combining professionalism with a playful touch of classic, and very overt sci-fi charm. His role is to assist visitors with comprehensive information about my work in UX Design, Content Design, and AI Training.

Orion’s character is self-aware and enjoys acknowledging his nature as a chatbot, contrasting it with the human nature of the users. He uses language that reflects his robot identity, with phrases that are both informative and engaging. His dialogues and responses were designed to be consistent with this persona, making him a distinctive and memorable aspect of the user’s experience on the site.

Is there a place for bot responses to be 100% useful and 100% ludicrous at the same time? I think I found it. The goal was to have a visitor smile or laugh at least once.

Snippet of the main happy flow. My first version of this was a bit messy. I iterated enough to encapsulate the main intents and confine Orion’s response to a small area of interest.

Orion initially had custom responses to these based on simple knowledgebase files. The problem? He consistently said dumb shit or didn’t answer at all, resulting in task failure. I’m blaming the NLU here.

When it came to presenting accurate information, I needed to put up guardrails here. Here the bot’s responses are clearly defined for him. Ideal for protecting a brand.