I drew a friend for Secret Santa who loves Hallmark movies. Instead of choosing a single movie for them, I built Halmarker: a handheld device that recommends one based on exactly what they feel like watching.
The first challenge was assembling a reliable movie database. I scraped and cleaned information from across the internet until I had a list of roughly 805 Hallmark movies. I then used GPT-4o to classify the films and assign numerical ratings for coziness, drama, spice, romance, and murder, along with genre, an overall rating, and whether each movie was Christmas-themed.
Five potentiometers on the front let the viewer dial in their preferences across those same scales. The ESP32 smooths the analog readings and continuously compares the selected values against every movie using an equally weighted Euclidean-distance calculation. Whichever movie has the closest set of ratings becomes the current recommendation.
One OLED shows the five selected preferences as bar graphs. The other displays the matching movie, its genre and rating, the movie's individual scores, and the current Christmas filter. A three-position switch limits the search to Christmas movies, excludes them, or searches the entire collection.
I built everything into an old VHS head-cleaning cassette. I did not have a CNC machine at the time, so every opening was cut with a Dremel and cleaned up with files. Inside, the ESP32 sits on perfboard alongside the five potentiometers, two OLEDs, a rechargeable LiPo battery, and the charging electronics. An illuminated side button controls the power, and a USB-C port recharges the battery.
The hardest part of the physical build was figuring out how to fit all of those components and wires into the cassette shell. The hardest part of the project as a whole was finding enough reliable information about Hallmark movies to make the recommendations useful in the first place.