you may pursue any of these, but please let me know if you do
Neovim LSP and successor to Roraer then a "shell" built from there (proportional font, Dotsies?, reloadable sections, a tad like ipynb) could make GUI operation to merge commands as by && or ;
automated drawing (photo/general image to image of a few curves), find or invent an algorithm candidate, beautiful-if-workable: binary space partitioning draw a line that cuts the image in half, then adjust/reshape it (control points?) to sharpen the average colour distance between the two "halves" then recurse on each partition; the divisions are the lines of the drawing tested it a little, doesn't seem to work
Anki export notes to PDF (look for prior art, especially built-in perhaps that i haven't found)
convert digit strings to words by mnemonic major system and a search thru Hebrew shoreshim (seems better than English on average for three- or four-consonant strings)
configure Corsair keyboard lighting, dim except for a bright flash when pressing each key
machine-learned program to identify orientation of an image
comments as a directed graph, rather than a tree must it be acyclic? let's find out
implement infix-only language further than ever before, with Prolog type inference (if that turns out practical) and compilation thru qbe see heylang notes file
highly configurable/reprogrammable e-paper smartwatch (definitely seek prior art) input by a small number of side buttons, and/or accelerometre/gyrometre for a Dasher-like text input method perhaps port a vi-like editor adapted for unusually few buttons, treating a repeat tap as more expensive than a different-button tap
ergonomically split-tilted keyboard, made cheaper and more flexible (figuratively and literally?) with capacitative touch-sensitive sheets instead of mechanical switches, maybe a couple conventional buttons under the thumbs
smart pen attached by stretch-sensing cables (three of them?) to fixed points, such that moving the pen stretches/compresses the cables such that a controller than infer pen-tip's position thus a drawing tablet: x and y of the pen transmit to x and y on the screen, when and only when z of the pen is close enough to z of the page two cables might suffice, if the pen is conductive as to enable a simple on/off-canvas test
sensor to affix to the neck to detect and report/deter head tremors
plain TODO list synced between devices, maybe Syncthing over a directory of numbered single-line files, but ideally find existing app
dimensional analysis practise problems, to get good at physics heuristics