This year at CampIO, we celebrated the culture of innovation as our employees shared personal projects, bringing us everything from an RPA “funland” to an AI cocktail generator.
We believe the best technologists are lifelong learners.
One of our favorite outlets for learning and sharing all things tech: CampIO, our annual internal technology conference. Join us as we revisit past iterations and recap the major highlights from this year’s event. From mixed reality VR experiences to beer goggles, stereo-integrated tube amps to AI-optimized marathon training, Legos on the cheap to Arduino-driven paintball turrets, why do we do it?
Year after year, tech professionals, enthusiasts and newbies alike come together to present their passion projects. In lieu of client demos or work-related project development, CampIO offers an outlet for the fun, off-the-wall projects our colleagues do on the side. But make no mistake – these lighthearted and experimental projects are where true innovation happens.
Technologists spend most of their time at work delivering. But delivering with value largely relies on the ability to explore new possibilities and reinvigorate our practices. In fact, CampIO is a major avenue for opening conversations around new and emerging technologies at Centric Consulting.
Looking back, blockchain, machine learning (ML), Ruby, clean architecture, chatbots and more were all discussed at CampIO – long before becoming integral to broader conversations regularly held here. As CampIO host Eric Galluzzo reminded us, you can change how Centric approaches new and emerging tech by participating in CampIO. We can also reuse presentations at other meetups, seeding new ideas around tech topics and inspiring clients and other technologists worldwide.
Building a 100-year company requires building 100 years of viability, far beyond our current resources and capabilities. On a much smaller scale, sustaining 100 years of CampIO success requires building in continuous adaptation.
We’ve experimented with a variety of formats over the years, including keynote speakers, client presentations, competitions between teams, and concurrent hackathons. As our needs and interests evolve over time, we’ll respond with continued experimentation to improve the CampIO experience for everyone.
When Eric asked other participants to reflect on why we have CampIO, the responses were rooted in our culture. A culture of vision, adaptability, engagement and connection. As one participant reflected, it’s about much more than remaining relevant in the face of disruption – events like CampIO directly contribute to our culture – and enable us to create the kind of company we hope our kids will want to work for someday.
That’s why when we say CampIO is open to everyone, regardless of technological prowess, we mean it. If you have an exciting idea percolating in the back of your mind, now is the time to take that inspiration to the next level. Who knows, you could be joining your friends in presenting at next year’s event – perhaps even with a return to an in-person format.
This Year’s Inspiration
Let’s highlight some of the key takeaways from our CampIO 2022 presentations. The too long; didn’t read version? Each presenter was uniquely innovative and worthy of their own spotlight.
AI Cocktail Generator
Senior Architect and former bartender Zach Tesler kicked off CampIO 2022 with an ambitiously boozy project. His goal? To build an AI cocktail generator in support of his quest for unique new libations.
First, Zach identified a solid source for capturing cocktail data. He scraped an online repository of more than 8,000 cocktails, ignoring preparation instructions to get to the heart of his quest – pure ingredients. He used UiPath RPA to then cleanse the data and prepare for ingestion, and he curated and analyzed cocktail nodes, ingredient nodes and added part-of relationships.
Adding affinity relationships was where things got interesting. New relationships demonstrated the potential of adding two or more ingredients that occurred somewhere else together in the data. Random matches led to the discovery of some unexpected matches that only the most daring cocktail slingers would attempt to create in real time.
However, Zach soon found that he needed to fine tune and train the base model furth – completed prompts led to somewhat predictable cocktail combinations over time, so he opted to provide around 3,000 samples of ideal output cocktails from the data set, run it again and create a new, more volatile and creative model. While it’s still a bit of a work in progress, Zach is confident his AI cocktail generator will soon deliver dramatically cooler cocktail results.
Cross-Platform App Development
Eric followed up the next day with a demo only somewhat cheekily referred to as “the relative mediocrity of cross-platform app development.” His challenge? IOS Swift, React Native and Android Flutter UI are all making cross-platform development easier – but he needed to test and evaluate this on his own terms.
Eric’s rubric for evaluating relative mediocrity included scoring across key elements like language, syntax, widget selection, components, tweak-ability, UI concerns like look and feel, layout, common tasks, asynchrony, persistent state, supportability, debugging, testability and general popularity.
During his demo, Eric found that React Native allowed him to flip between IOS and Android versions of UI demo screens with relative ease. After switching to Flutter, the Android screen looked like an Android screen – relative success. However, Flutter still looked rather Android-like on an IOS screen – a demerit.
Eric eventually concluded that you can’t escape that Google look and feel and awarded React Native as the slightly easier tool for creating cross-platform UIs with a good look and feel. Even still, Flutter offers a nice dev experience with tooling, DartPad, a great library and solid documentation – if you can get past the constraint systems and verbosity.
ML Chord Prediction
We also enjoyed a tuneful presentation given by Jeff Kanel, who plays fiddle in an Irish folk band when he’s not leading our Data and Analytics practice. When Jeff’s guitarist came to him and said, “Where are the bleeping chords?” regarding a traditional tune about a gal named Maggie, Jeff had a hunch. If trained musicians can create the ideal chords for any given song’s notation, why can’t we train ML to predict chords, too?
Similarly to Zach’s cocktail project, Jeff started with data acquisition, scraping 36,000 tunes from the web to then load, interpret, convert from legacy music modalities to more common, modern keys, generate tune metadata, features and extracts. Around 30,000 tunes had no chords at all – and the remaining had some – but were they any good?
Jeff’s landed on over 750, 000 training sets, and then model training came next. Azure ML, a multiclass neural network, gave a 99.5 percent accuracy rating to the model. But as Jeff was quick to point out, anytime we see that high an accuracy rate, there’s likely something off on the dev end.
He found data leakage to be the culprit – perhaps he was feeding some chords inadvertently through some of his training sets. After retraining and rerunning his model, our musically minded technologist got a 65 percent accuracy rating – too low. While Jeff had believed that musicians should be able to intuitively figure out similar results for chord prediction, all the data proved otherwise. Everyone online appeared to be finding different chords for the same song – and the result for Jeff’s project was a cacophony.
So how did his ML do on the original tune in question? Lots of dissonance, too many chords and even the original training data (ground truth) may have been of such poor quality that those chords, when demoed live during the presentation, weren’t placed correctly in the eyes of our music theorists.
Would post-processing help? Cleaning up the data? Another participant suggested training the ML on not just Irish folk tunes but common chord progressions from other genres to help fine tune its training. As much a technical skill as it is an act of artistry, Jeff will likely continue to work on his ML chord predictor – or at the very least, encourage his guitarist to get a bit more creative.
Our resident Disney expert and Senior Data and Analysis Manager Becky Gandillon was awarded the Dingey Yellow Book Award for Most Useful Presentation. If you’re curious about that yellow book, it’s the classic “The Mythical Man-Month: Essays on Software Engineering” by Fred Brooks. If you’re curious about Becky’s useful presentation, just consider her your go-to source for a data-optimized family vacation.
Becky is a self-avowed super planner who relies on data to curate the perfect Disney experience. She sees value in using Disney problems as a framework for approaching other data problems. Becky started off with some simple goals in mind: don’t analyze data for data’s sake alone, avoid crowds, save money and enjoy your vacation.
The challenges? There were many, including capacity limits, variable ticket costs, variable attendance, operating hours, four parks – each with up to 39 attractions, 30 or more dining venues, plus 34 resorts, two waterparks, numerous up-seller opportunities and individual needs and preferences.
All this created a big optimization problem, but Becky put predictive data analysis to the test and developed a crowd calendar of dates, resorts, prices and weighted crowd levels – all on a numbered scale based on time of wait. She also scraped Disney menu data to find to most affordable yet delicious dining options plotted to a graph for targeting reservations, created a heat map along with confidence intervals to determine sell-out likelihood on lightning reservations to key attractions, and much more. With the potential to save hundreds of minutes of wait time per day – and thousands of dollars per trip – Becky’s data-optimized approach to vacationing was a clear winner.
Becky also ensured her presentation was highly interactive, prompting further discussion of adding inputs and affinities for weather, age or height restrictions, relative like or dislike of certain costumed characters and more. In sum? Even the most Disney-jaded are likely to call on Becky next time they’re on the hook for planning a family vacation.
Not to be overlooked, two resident robotic process automation (RPA) experts, UiPath MVP Tracy Dixon and Senior Consultant James Taylor, treated us to a dual presentation. These RPA power players gave us a sneak peek into how they envision delivery excellence and project booking enablement via RPA as a “funland” for tech consultants.
As many of us already know, every project comes wrapped up with a whole host of data and CRMs that must “talk to each other” to deliver at our very best. RPA has changed significantly just in the last year or so, with a much larger focus on data integration leading to a new integration service that launched earlier this year. This service is a library of feature projects allowing users to securely log in and connect features, load and save projects, and share configuration and customization of new activities within any given project across all relevant features.
While most non-technically minded folks are at least somewhat familiar with the concept of RPA by now, some still think it’s in the realm of actual robots taking out the trash. While this sounds a little too much like the Jetsons to some, the reality is not that far off. Physical RPA is possible when your task connects to the IoT – meanwhile, we consultants can share the benefits of delivery excellence RPA with our clients.
Our week of presentations wrapped up with none other than this year’s Most Innovative winner, Jeff Aalto. This Senior Architect is on his way to becoming a master of smoked meats and decided it was time to up his game with a little Raspberry Pi. Essentially, Jeff’s goal was to create an app that utilizes live data to optimize smoking temperatures and times against current weather conditions and particular smoked meat dishes.
His objectives? The result of smoking is delicious – but it takes far too long. Serious smokers need real-time data to go about their day-to-day tasks. Jeff opted to use push data (food, weather report visuals) and big data (historical weather data, report visuals) to physically do something (via Servo controlled vents), to receive temperature threshold alerts, to analyze and do it all in a workflow – and to do it all as cheaply as possible.
A tall order, perhaps, but not so for Jeff. Despite running nearly two years’ worth of data from the cloud to his local environment (and running up a $600 bill), MSAL authentication denial, SD card corruption, network hacking attempts from Belgium, and limited soldering prowess resulting in burnt chips (not the potato kind), Jeff landed on a highly functional outcome.
He treated us to a mouth-watering live demo of smoked sausages, served by real-time food cooking temperatures, near-real-time weather forecasts, text alerts at peak readiness, and Servo-driven vents to allow perfectly optimized airflow – directly attached to the smoker via Jeff’s dad’s old Erector set.
Jeff can also overlay the cook data he collects to capture a full history of events, as well as a cookbook dashboard for determining the optimal targets according to each dish’s recipe.
After all of this experimentation – and five weeks of solid smoked meats – Jeff is growing weary of eating the results of this project. Still, his goal is to acquire more data to help perfect his smoking skills over time. Recipe ratings are simply one tantalizing data prospect left to acquire. If you’d like to help Jeff test and rate his menu, tell him CampIO sent you.
While that’s a wrap for this year’s CampIO event, we hope it encourages you to get creative and share your backyard, garage or basement passion projects with us all. Remember, you don’t need to be a seasoned tech expert – success lies in your innovation and daringness to try.