Description: I designed and developed a suite of Alexa skills that generated information on various topics from movies and sports, to food and fitness.
Features
Comparative analysis in a conversational context
Responses generated from parsed API calls
Multiple intent options in the skill greeting
Request clarification and disambiguation
Experience
I began experimenting with Alexa skills when the first developer kit dropped in 2016. By 2017 I was developing Alexa skills powered by NLP and NLU models that enhanced the conversational abilities.
One skill imparticular was called the Autocompare skill which allowed real-time comparative analysis between two entities on various topics. For example using built-in intents for food and movies I was able to use API calls to instantly gather data on the requested entities, feed them to a custom model and generate responses in comparative format years before the wide spread release of LLMs.
Asking for a comparision within a specific topic (e.g., foods, movies, products) was tricky in a graphical UI and downright difficult in a voice interface. I used the initial greeting to introduce early users to how to craft their prompts and created informative error messages to prevent behaviors that attempted to request cross topic comparisons.
After conducting various demonstrations at tech meet-ups and gathering user feedback I published the skill on the marketplace.
The conversational flow allowed users to list the things they wanted to compare more seamlessly than a web interface. Users expressed satisfaction with the broad range of foods that the Alexa skill recognized but were not satisfied with the latency of the responses. Overtime the parsing functions became more efficient and response times improved from 5secs to 1 second.
Design and Development
My initial tests with early users indicated that 2 or more requests were typical in one session. This lead to incorporating follow-up requests in the conversational flow.
To accomplish this I designed the backend architecture to support the skill with additional intent recognition and serverless functions to fetch and parse data as fast as possible for low cost.
The Amazon provided templates in 2017 enabled simple functionality however, I was able to deliver a deeper level of conversational experience by leveraging my knowledge of AWS.
–