MSc Dissertation Project
Subcept
An adaptive AI tutoring engine built for my MSc Computer Science dissertation at the University of Birmingham. Rather than handing learners an answer, Subcept turns their question into a structured lesson, teaches it through conversation, and diagnoses why an answer was wrong.
- Generates a complete lesson plan for each question using an LLM, with sub-skills, worked examples, marking rubrics and common misconceptions, validated before it is served
- Tracks each learner's mastery of every sub-skill with Bayesian Knowledge Tracing, updated from rubric-graded answers
- Classifies wrong answers by failure mode and responds with matched remediation, from revisiting a prerequisite to a worked example
- Built-in guardrails, screening questions before generation and preventing the tutor from simply giving the answer away
- Controller-Service-Repository architecture tested with xUnit, containerised with Docker and deployed to a Linux VPS
C#ASP.NET CoreEntity Framework CoreSQLiteReactTypeScriptTailwindGemini APIxUnitDocker









