Your textbook,
but it talks back.
We built an AI tutor that actually knows your course materials. Not generic ChatGPT. Course-specific RAG with real citations, IRAC structure, and answers grounded in your actual textbook.
40% of AI legal citations are fabricated.
Law students are already using ChatGPT. The problem is it doesn't know their textbook, their assessment format, or Australian case law. It invents citations. It ignores IRAC structure. It gives confident answers grounded in nothing.
Generic AI tools have no context about course-specific requirements. A student studying Remedies at Macquarie gets the same response as someone studying Torts at Harvard. The content isn't wrong per se. It's just not relevant.
The gap was clear: students needed an AI tutor that could search their actual textbook, reference specific pages, and produce structured answers their assessors would recognise.
RAG that reads your textbook before it answers.
Three answer modes for different assessment types
IRAC mode for problem questions. Essay mode with AGLC4 citations for academic writing. General mode for tutorial prep and concept review. Each mode has dedicated system prompts that enforce the right structure.
Students pick their mode, select their topic week, and get responses grounded in the exact chapters they're studying. No irrelevant tangents. No hallucinated cases.
Every citation is real
Textbook content ingested as structured markdown with semantic chunking. Each response references specific sections and page numbers (e.g. “Remedies, S1.1, p. 4”). Closed-book constraint means the AI can't cite what isn't in the textbook.
Three models, no lock-in
GPT-5-mini for standard queries. Gemini Flash for speed. Claude Sonnet for deep analysis. Credit-based pricing lets students choose the right model for each question. Vercel AI SDK abstraction means swapping providers takes zero code changes.
Production-ready with 200+ legal concepts indexed.
Course Companion launched with 2 complete law courses from Macquarie University (LAWS5000 and LAWS5056), covering 17 distinct topics across full semester content. Over 200 discrete legal concepts are indexed and searchable.
First-token latency sits under 2 seconds. Semantic search retrieves relevant content in under 500ms. Citation accuracy is 100% by design: the closed-book constraint means every reference traces back to the actual textbook.
The multi-model architecture means no provider lock-in. When new models launch, they slot in without rewriting the chat pipeline. Students pick the model that fits their question and budget.
Common questions about AI tutoring platforms
Have an AI product idea? We build them.
Course Companion is one of our ventures. We build AI-powered products from concept to production using the same stack and speed we bring to client work. Got a concept? Let's talk.