Engrammo
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Memory champions don't have better brains. They walk through a place they already know. Engrammo uses Muse Spark, SAM 3.1 and Muse Image to turn a photo of your own room into that place in a few minutes, and then checks whether it worked on you. Why: students and professionals forget most of what they study within days, and brain-training games mostly improve the game itself. The method of loci is a different kind of tool: a strategy for the material you actually need to learn, and one of the best-studied memory techniques there is. Almost nobody uses it, because building a good palace by hand is slow and abstract. Engrammo does the hard part, on your room and your material: Snap your room. Muse Spark (vision) finds the landmarks and puts them in walking order. SAM 3.1 cuts each one's exact silhouette out of your photo. Add what you need to learn: type it, pull it from your goals, or photograph your handwritten notes and Muse Spark reads them into a list. Muse Spark ties each fact to one object with a vivid, absurd link ("the Estates-General: three beanstalks in velvet hats sprouting from your plant pot"), and Muse Image paints that scene. Recall: the silhouettes become the cues. You walk the room in your head and type what lives at each spot. Muse Spark grades you like a tutor, not a string matcher, so a near-miss spelling still counts. Check it on yourself: Engrammo quizzes you once without the palace and again with it, and shows both scores side by side (one learner, not a study). "Prove it" hides the photo for a blind run, and a spaced-repetition scheduler brings the palace back just as you start to forget. It also has a Muse Spark Session Coach agent that plans your day through tool calls, a lesson curriculum that walks your own palace, and full English and Hungarian support. Every AI step is visible: "What the AI did" lists each Meta call behind your palace. Reviewers without an API key can watch a clearly labelled replay of a real recorded run, with genuine model output. Built solo during the hackathon (Oct 27 – Nov 4) in Flutter, with CI and 1,242 automated tests.