Day 15 - Working on the go
Yesterday I left Tokyo for a group trip to Ito.
For the past 1-2 days I have been thinking of the recommender system for the knowledge graph and I arrived at a v0.1 and some future iterations to test thereafter.
Before starting the trip, I tired to structure myself so that I can do thinking activities on my mobile while we are out and about. When we all stop somewhere - that's when I can 'work'.
It's funny how yesterday I wrote that I didn't need a computer to 'think'. And while that's true, the surface area a computer gives me helps me think better.
For the recommender system, I started small. Each card has different swipe options. Here is the basic logic for two things I want to start with:
5.1.1 "Already Know This" → Level Progression
"already know this" → difficulty too low, skip ahead
Detection Threshold
Flag after 3-5 "already know this" actions within a single week:
- 3/21 cards (14%) = possible signal
- 5/21 cards (24%) = strong signal they're beyond this level
- If clustered around specific concepts = they know that concept, not necessarily everything
Response Strategy: "Spot Check, Don't Skip"
Week N: User marks 5+ cards "already know this"
Week N+1 recommendation:
- Keep same goal active (maintain storyline continuity)
- Replace 30-40% of Week N+1's cards with "probe cards" from 2-3 weeks ahead
- Keep probe cards thematically aligned with Week N+1's theme
- Track performance on probe cards
This process is very similar to my process before to build out the spec. I build it -> feed it to Claude and Gemini to identify inconsistencies - read through their feedback -> reiterate till I feel confident in what I have written.
