3 Lessons Learned in Building a Production AI Platform from Delta Bravo
Modern Devs Charlotte hosts Delta Bravo team for presentation, live ML coding exercise
Delta Bravo‘s John McAliley and Rick Oppedisano presented to a packed house this week for Modern Devs Charlotte‘s monthly meetup at Wray Ward in uptown Charlotte.
“There’s a ton of interest in AI and Machine Learning,” says Oppedisano, “we want to help people understand where they can start and also learn from our journey. The lessons we’ve learned along the way could save them months of technology evaluation and development time and in some cases, thousands of dollars in system costs.”
Start Small with Familiar Tech
The presentation started with some context of how and why the Delta Bravo AI platform was built. This helped articulate a clear AI use case and also defined the differences between data science, machine learning and artificial intelligence. “Determining if you have an AI use case at all is important,” says Oppedisano. “The ability to clarify that use case and establish the measurables behind it is the first step in the process.”
Once the use case is established, the next step is prototyping. The Delta Bravo team recommends to start small and with a technology you’re already comfortable with. “There are a ton of choices out there and it can be a bit overwhelming,” says McAliley. “There’s likely an option in a tech stack that you’re already working with. Start there to save time and focus on a small, controllable and quickly proven use case.”
Sample Use Case, Live Machine Learning Code Exercise
McAliley followed up with a sample AI use case based on Image Recognition and a live Machine Learning code demo designed to show the differences in accuracy based on data quality and model construction.
A model was established based on metrics that were clearly and visibly defined. “Accuracy is synonymous with credibility,” explained McAliley. “This exercise shows how to get the highest degrees of accuracy, as well as where certain compromises exist and where you begin to get diminishing returns on your model.”
The team also shared Delta Bravo’s functional architecture and provide an example of how AI is used within the platform to recognize, quantify and predict an oncoming database performance problem. This helped the audience visualize the differences in and applications of data science (insight), machine learning (prediction) and artificial intelligence (recommended action).
Lessons Learned
Oppedisano and McAliley shared 3 Lessons Learned from their experience, ranging from technical choices to the importance of data. “The quality of the data you are putting into the model is at the heart of everything,” says Oppedisano. “It’s a classic garbage in, garbage out scenario.” The presentation covered some of the more efficient options available for data cleansing and grouping, as well as the importance of representing the data in a clear, visible manner. “In certain scenarios, the AI process presents the data more efficiently than a human,” says McAliley. “Being able to visualize the data itself, then drill down into insight, correlation and action is important.”
Oppedisano and McAliley shared a fourth “Bonus Lesson” about understanding and managing cost projection. “Machine learning requires a lot of processing,” says Oppedisano. “Its easy for cost to start becoming a factor.”
Feedback and Discussion
“Attendee feedback was very positive,” says Oppedisano, “but the real impression we got was how much opportunity there is in this space.” Several potential use cases and solutions were discussed during and after the presentation, with a lively discussion between the attendees and presenters.
“There’s a lot of passion in the AI community at large,” says Oppedisano. “We want to be a part of bringing that passion and knowledge to the local technology community.”
BOOK US FOR YOUR MEETUP.
Interested in topics like this and looking for qualified, professional and fun presenters? We’d be happy to help! Click below to connect with us.
