Fluxor

Inspiration

E-commerce platforms have faced persistently high checkout abandonment rates for years, with much of the drop-off caused by small but impactful UX friction. While teams collect large amounts of behavioral data, turning that data into clear, timely design decisions remains slow and manual. We were inspired by the question: what if interfaces could learn and adapt on their own, instead of relying on endless redesigns and A/B tests?

What it does

Fluxor is an adaptive UX system that observes real user behavior, detects friction points, and automatically adjusts the interface to improve retention. By analyzing behavioral signals like hesitation, drop-offs, and repeated actions, Fluxor generates targeted UI changes, tests them in parallel, and keeps the version that performs best. The product’s identity stays intact, while the interface continuously evolves around how users actually behave.

How we built it

We built a simulated checkout flow instrumented with Amplitude-style event tracking. User interactions are logged as behavioral events and passed into an AI layer that clusters patterns and infers likely UX friction. The AI selects adjustment strategies such as simplifying steps or adding guidance, deploys them as UI variants, and evaluates outcomes using retention metrics.

Challenges we ran into

One major challenge was avoiding a system that felt like a simple rules engine. We focused on having the AI infer why users struggle, rather than reacting to single thresholds. Another challenge was scoping: building a full design editor wasn’t realistic in a weekend, so we constrained changes to meaningful micro-adjustments that clearly demonstrated the concept.

Accomplishments that we're proud of

We successfully demonstrated a closed-loop system where user behavior drives AI decisions, and analytics validate those decisions over time. We’re proud that Fluxor doesn’t just measure UX problems, it actively responds to them. We also built a demo that clearly shows how even small improvements in retention can compound into meaningful impact.

What we learned

We learned that analytics alone don’t solve UX problems. Decision-making is the real bottleneck. We also learned how important it is to clearly define the boundary between automation and control: adaptive systems work best when they respect product identity rather than constantly reinventing it.

What's next for Fluxor

Next, we’d expand Fluxor beyond checkout flows to onboarding, navigation, and content-heavy experiences. We’d also introduce stronger learning over time, allowing Fluxor to build product-specific intuition about what kinds of adjustments work best. Long-term, Fluxor could integrate directly into existing design and analytics workflows as a self-improving UX layer.