I'm Ashwini! Founder and CEO of Vue.ai. AMA!

Hey Krish! Glad to be here. The question is fairly broad, so super high level answer here, more principles and guidelines than anything else. If you had specific areas you’re looking for answers to, holler!

  1. KNOW YOUR DATA

Your data = people or the things they use or content they interact with or goods they buy. Your data is subjects with stories and experiences before it’s statistics. And before you embark on any journey to build AI enabled use cases or automate workflows, the first thing to do is to ‘know your data’. It’s important that you understand your data, your subject before you decide to automate. All our product people, AI teams, customer success teams are trained on interpreting data before they build or talk to customers. How you train your data will depend on the patterns and inferences and assumptions you make, so this is very important.

  1. MAKE YOUR DATA TEACH YOU, ASK QUESTIONS

Just like 1, its important to cut the data a bunch of different ways by asking several questions, relating to the workflow you want to automate or the use case you want to integrate AI with in some way. You’ll be surprised at the patterns you discover. Big part of the research process is to let your data sing :slight_smile: and then understand it in the context of the subjective / anecdotal / qualitative observations you do.

  1. HYPOTHESIZE, EXPERIMENT, ITERATE

Now you’re ready to use the data to train new models for whatever use cases you’re looking for. Like with any new feature building, experimentation, iteration is super important. There’s so much hype around AI. If your AI is not adding meaningful value to your customers, don’t do it. Have a hypothesis on what you want your AI to do, test and evaluate with customers and focus on the data emerging. You can setup several experiments in parallel with different control groups and test new ideas. Evaluating usefulness or value of AI features is a topic I’ve been interested in for years now. Depending on whether you’re simply looking to automate / increase speed / throughput vs. trying to change human behavior, the outcome can happen as quickly as immediately or can take as much as years to have an impact.

4 Likes