WonkaLift

Exercise Selection
Workout Page
End of Workout
Gameplay
Profile Page
On wrist
Standalone

Inspiration

Every single year, people set all kinds of goals for themselves — professional, personal, relationships. But let's be honest: the vast majority of them come back to fitness. In fact, a survey found that 9 in 10 Americans set fitness goals at the start of the year, yet most struggle to keep up with them within just a few weeks — sometimes even days.

After some research, we identified two core reasons why this happens. The first is the lack of immediate visible progress, which causes a sharp drop in motivation. The second is the pressure to stay consistent, which eventually gets the best of most people. Despite the countless fitness products already on the market — tracking calories, heart rate, workouts — none of them actually address these two root causes of quitting.

So we did.

What It Does

Unlike other projects that attempt to tackle these problems, WonkaLift uses no cameras whatsoever — because let's be real, nobody in the real world is going to pull out their laptop or prop up their phone just to film themselves lifting.

Instead, the user wears a custom-designed wristband that uses an accelerometer and gyroscope to perform three core tasks: count reps, ensure proper form, and maintain consistent tempo. Dozens of scientific papers correlate these three factors with increased muscle growth — so we built around what actually works.

The wristband streams this data over Bluetooth to the user's phone, which does two things. First, it tracks your workout data over time. Second — and this is the fun part — it gamifies the entire experience.

In keeping with Hacklytics' Willy Wonka theme, we built a game where Wonka travels along the chocolate river in his boat. Every rep is analyzed in real time: a good-quality rep keeps the boat sailing smoothly, while a bad one sends it crashing into the walls. Too many crashes and you lose. But finish your set with 80%+ good form and tempo, and you earn a Golden Ticket. 🎫

Beyond the game, users can share their progress and compete with friends or family to stay accountable. We also provide a projected strength progression curve so users can see how much stronger they'll get over the coming weeks — keeping them motivated and driven to keep showing up.

How We Built It

WonkaLift has two main components: hardware and software.

Hardware: The wristband is built around an ESP32 paired with an LSM9DS1 9-DOF IMU (accelerometer, magnetometer, and gyroscope). We collect acceleration data across all axes and use it to calculate rep counts and tempo metrics directly from raw IMU data.

Software: Raw IMU data is streamed to an iPhone over BLE. The app itself is built in React Native (bare workflow with Xcode, for reasons discussed below) and is designed to be intuitive and easy to use. Before usage, we collected 300 samples across the two most common dumbbell exercises — bicep curls and lateral raises — and used a 1D CNN trained in TensorFlow to classify each rep as good or bad. The trained model runs locally on-device, providing real-time inference with no noticeable latency.

This feedback is reflected live in the Wonka boat game. For strength projections, we use the Brzycki formula for 1RM estimation and apply a quality multiplier derived from the user's average form and tempo scores — meaning better training quality directly accelerates your projected progression curve. Our ML pipeline was built and trained using Databricks notebooks.

Each user has their own profile, with state managed via React Context. We also added an optional text-to-speech system that intermittently coaches the user in real time — reminding them to slow their tempo, hit full range of motion, and more.

Challenges We Ran Into

Compatibility issues were our biggest hurdle. BLE on the ESP32 doesn't play nicely with Expo Go, which would normally streamline React Native development — forcing us to switch to bare React Native with Xcode mid-hackathon.

On the ML side, training on only 300 self-recorded samples made our model highly susceptible to overfitting. Rather than simply re-recording the dataset, we applied targeted post-training techniques to improve accuracy. This included normalizing per sample rather than per rep, and building a correlation model between tempo and form classification — so a rep scored as "good form" but with an off-tempo could still receive a lower quality score. This helped us handle edge cases where binary form classification alone wasn't sufficient.

Accomplishments We're Proud Of

We're proud of identifying a meaningful gap in the fitness wearable industry — one that major players like Apple, Fitbit, and Garmin haven't addressed — and prototyping a viable, hardware-software solution in just 36 hours.

We're also proud of deliberately avoiding computer vision. We genuinely believe CV-based fitness tools aren't viable in the real world, and we're proud to have found a scientifically grounded workaround that actually works in practice.

What We Learned

Building WonkaLift gave us hands-on experience at the intersection of embedded systems, machine learning, data engineering, game development, and motion tracking. We learned how to deploy lightweight ML models on-device with no perceptible latency, and we saw firsthand how gamification can make something as demanding as consistent training genuinely enjoyable — while keeping it scientifically grounded.

Working with a small, self-collected dataset also taught us hard lessons about the challenges of limited data, and how thoughtful pre- and post-processing can meaningfully improve model quality.

What's Next for WonkaLift

Due to hackathon time constraints, we limited our exercise library to two dumbbell movements: bicep curls and lateral raises. Moving forward, we want to expand to a much broader set of exercises, and build personalization into the model so it adapts to each individual user's goals over time.

On the hardware side, we'd like to replace the breadboard with a custom PCB to shrink the form factor into something truly wearable. We're also exploring adding more mini-games to keep the experience fresh, and — most ambitiously — bringing WonkaLift into AR/VR, where the game, rep count, form score, and tempo all overlay your workout in real time.

The Golden Ticket is just the beginning.