Chief of Learning.
A Mandarin learning system that adjusts the next study plan using real activity, feedback, and long-term goals.
IN PLAIN ENGLISH
What this is.
Chief of Learning turns a long-term Mandarin goal into a daily plan. It uses study history and tutoring feedback to change tomorrow’s work instead of repeating generic lessons.
01 / THE PROBLEM
A streak is not a study plan.
Language apps kept activity moving but did not remember what I had studied or what I repeatedly got wrong. Chief of Learning takes on the planning and grading work so the student can concentrate on learning.
02 / HOW IT WORKS
One coach, three surfaces, one reasoning pipeline.
A Telegram chat, morning brief, and private dashboard share the same pipeline. Notion is the editable source of truth for lessons, corrections, and plans. A local recorder transcribes lessons before the text is filed back into the system. The next plan can respond to a miss instead of starting over.
03 / ENGINEERING DECISIONS
Derived search never outranks the source.
A semantic index in Neon Postgres with pgvector is a one-way copy of Notion and can fail open: if it is empty or unreachable, the system answers from Notion. Model roles are routed to the least expensive capable provider with fallbacks. Writes are crash-aware, transcripts are deduplicated by content hash, and failed pushes are replayed rather than dropped.
04 / WHERE IT STANDS
A working single-user system.
This is a working template for one person, not a public product. It carries 276 tests across 40 files. Lesson audio stays on the local machine and only its transcript is sent upward. The system is valuable because it now handles the daily plan and self-correction loop, not because it has a growth chart.
THE HONEST VERSION
What I am actually counting.
The honest boundary is single-user scope. It is a working personal system with tested seams, not a SaaS product waiting for a launch announcement.
TRY IT
Try the interaction.
TRY IT · STUDY CHECK-IN
TOOLS AND CONSTRAINTS
- Notion
- Neon + pgvector
- Telegram
- whisper.cpp
- Model routing