Inspiration
Memories don’t really disappear; they just fade when we stop noticing them. A lot of our best moments are already captured, photos, short videos, things we once cared enough to save. But over time, they get buried in camera rolls and cloud storage, rarely revisited. We wanted to change that. Not by asking people to search for memories, but by letting those moments come back naturally, in the spaces we live in every day. At its core, this project is about bringing memories back into view, at the right time, in a way that feels effortless. And while that can be meaningful for anyone, it can also quietly support people who struggle to recall moments by letting those memories resurface on their own.
What it does
Our system turns your living space into a dynamic memory display. It captures meaningful moments, selects the most aesthetically pleasing frames, and displays them when you’re actually there to see them. When you walk by and linger, the system recognizes your presence and intention, then brings those memories to life on the screen. It’s not just passive playback. It responds to you. Even better, it can go beyond connecting with your friends. They can send photos or videos directly to your space, turning the display into a shared window of memories. Instead of messages getting lost in apps, they appear somewhere you see every day. It’s less like a device, and more like a living photo frame that understands when a moment matters.
How we built it
We built the system around three main components working together: A transmitting module powered by a Rubik Pi handles incoming media and runs a lightweight AI model that selects the most visually appealing and display-worthy frames from videos. A central controller using an Espressif Systems ESP32 S3 runs edge intelligence. It combines computer vision from ESP32 S3 Sense for face alignment detection with signals from ultrasonic and PIR sensors to understand when someone is present and intentionally viewing the display. Instead of reacting to any motion, it predicts “lingering” behavior, meaning the user actually wants to look. Finally, a display module (another Rubik Pi) renders the selected images and videos on screen while also hosting a lightweight server. This allows seamless communication between devices, including uploading new content and selecting what gets shown. Together, these pieces create a system that senses, selects, and displays memories in real time.
Challenges we ran into
Displaying video turned out to be much harder than expected. The ESP32 is great for sensing and lightweight edge AI, but it doesn’t have the storage or power to handle video playback, especially without onboard SD support. Introducing a Pi solved that, but added complexity in communication, synchronization, and system setup. Another challenge was training the image selection model. Picking a “good” frame isn’t just about objective quality; it’s subjective. What feels meaningful or aesthetic to a person isn’t always obvious to a model. We had to build our own dataset and experiment with ways to balance visual clarity with emotional value.
Accomplishments that we’re proud of
We’re proud that the system doesn’t just work, it feels right. When someone walks by, pauses, and the display responds with a moment that actually resonates, it creates a small but powerful experience. It turns a screen into something personal. Beyond that, we built a full pipeline from sensing human behavior to selecting meaningful content to displaying it in a natural way. It’s not just hardware or AI, it’s the interaction between them that makes it special. And while it wasn’t our main focus, we love that this could also support people with memory challenges by bringing familiar faces and moments back into their daily environment without requiring effort.
What we learned
We learned that timing matters just as much as content. A beautiful memory shown at the wrong time feels random. The same memory shown when someone is actually present and looking feels intentional. Bridging that gap between human behavior and system response was one of the most important parts of the project. We also learned how to split intelligence across devices. Lightweight sensing and decision-making on the edge, heavier processing on more capable hardware. Getting that balance right made the system both responsive and capable.
What’s next?
Next, we want to bring the social side of this idea fully to life. We’re working toward making friend-to-friend connections seamless, so people can send moments directly into each other’s spaces in a way that feels immediate and personal. At the same time, we’re improving the sensing and prediction side of the system. Better accuracy in detecting when someone is truly present and engaged will make the display feel more natural, less like a trigger, and more like something that understands you. We also want to refine how memories are shown. Instead of a single frame, we’re exploring ways to display multiple moments together in a clean, non-overlapping way, so the experience feels more like reliving a memory rather than just seeing a snapshot. Because at the end of the day, the goal isn’t just to store memories. It’s to help you live with them again.
Log in or sign up for Devpost to join the conversation.