Our philosophy

Alongside school

School and Subsi do different jobs.

School brings human teachers, classmates, collaboration, community, physical materials, and hands-on experiences. Software should not pretend to reproduce those relationships.

Subsi adds the one-to-one layer: time to stay with one question, explain it another way, give feedback on the learner’s work, and move across subjects without another appointment.

The Subsi thesis

TeachPractiseCheck

For this learnerWhat they knowWhat interests themWhat their work shows

What we believe

Six principles shape every lesson.

1. 01

   Give them a path, not a firehose.

   A Curriculum or Course supplies a useful route. Subsi chooses the next step without closing off questions or curiosity.

2. 02

   Start from this learner.

   One-to-one teaching begins with prior knowledge, age, pace, and context. The explanation changes; the truth and learning goal do not.

3. 03

   Teach—do not just answer.

   Subsi chooses the useful question, explanation, example, visual, or hint—then hands the thinking back.

4. 04

   Ask the learner to produce.

   Explaining, solving, drawing, comparing, and retrieving reveal more than reading a fluent answer ever can.

5. 05

   Let feedback change what comes next.

   A useful check leads to specific feedback, a repaired foundation, another explanation, or a new challenge.

6. 06

   Use interests with purpose.

   Interests shape examples when they genuinely help; they never replace the capability the learner is meant to understand.

Research behind Subsi

Three findings that shaped how Subsi teaches.

They influenced how Subsi handles help, feedback, and independent work.

Sources

Key pieces behind our approach.

We link to the original paper, publisher, or research institution wherever possible.

• Carroll — A Model of School LearningTime needed can vary without changing the learning opportunity.
• Bloom — The 2 Sigma ProblemOne-to-one tutoring and a feedback-corrective loop are a historical challenge, not a promised modern effect size.
• Kalyuga, Ayres, Chandler & Sweller — The Expertise Reversal EffectSupport should reflect prior knowledge and fade as it becomes redundant.
• Bastani et al. — Generative AI without guardrails can harm learningSupported performance can rise while later independent performance falls.
• Kestin et al. — AI tutoring and active-learning physicsA promising undergraduate result built on deliberate instructional design.
• Wang et al. — Tutor CoPilotAI strengthened human tutors’ practice; it did not automate the human relationship.
• Stanford SCALE — The Evidence Base on AI in K–12A current overview of the causal evidence—and the questions still open.