A tutor learns to ask better questions.

A tutor that models the learner's situation — goals, deadlines, errors, doubt — not just the syllabus.

2025—26PRODUCT LEAD · FULL-STACKLLM ORCH. · NEXT.JS · MONGODB1,200+ SESSIONS · +38% RET. ※

THE SYSTEM UNDER TEST — SCATTER / REASSEMBLE · PROCEDURAL

THE 45-SECOND READ

PROBLEM

Every message was treated as a fresh start; the learner's context vanished between sessions.

DECISION

Make the learner's situation — not the chat log — the first-class object the tutor reasons over.

PRODUCT

A tutor that remembers goals, deadlines, errors, and doubt, states its understanding, and lets the learner correct it.

STATUS

1,200+ pilot sessions, retention +38% — draft figures pending verification.

THE TURN

A query became a situation.

The working case asks what changes when a tutor remembers goals, deadlines, errors, and doubt instead of treating each message as a fresh start.

THE PRACTICE / 01

Make the model answerable.

What the system believes about a learner should be inspectable, correctable, and subordinate to the learner's own account.

AI Tutor product still — 1600×1000

EVIDENCE / 02 — DROP REAL CAPTURES HERE

GENERATED IMAGERY IS NEVER USED AS EVIDENCE

Pilot methodology chart

FIG. 02 — PILOT METHODOLOGY

Verified outcome figures

FIG. 03 — VERIFIED OUTCOMES ※

EVIDENCE LEDGER / 03

What the working case shows.

— A running tutor prototype with an explicit learner-situation model— The situation stated back to the learner, open to correction— LLM orchestration on a Next.js / MongoDB stack

What this page does not claim yet.

— Session count and retention figures are draft placeholders, pending verification— Pilot methodology has not been published for review— Real product captures still need to be dropped into the figure slots above