Own Product · Cambridge O & A Level AI TutorLive: MVP Deployed

AcePrep AI

Snap a question. Get an examiner-style answer. Grounded in real past papers - not a generic chatbot.

15+
Cambridge Syllabuses
Sciences, Business, Humanities, Mathematics & Languages
3
Connected Study Loops
Answer a question, check my answer, live quiz checker
Grounded
Or It Doesn't Answer
Declines rather than inventing when the source can't support it

Client Challenge

Cambridge O & A Level exam boards publish past papers, mark schemes, and examiner reports as disconnected static PDFs - nothing connects them for a student in the moment they're stuck. Mark schemes use dense, technical language most students can't interpret alone, examiner reports (which carry the exact insight that explains lost marks) go largely unread, and private tutoring can fill the gap but varies widely in quality and price. Often the real gap isn't content - students know the material and still lose marks on technique.

Objectives

  • Connect past papers, mark schemes, and examiner reports into one grounded study loop.
  • Mark a student's own answers the same way an examiner would - awarded vs. total, strengths, missed points.
  • Never invent an answer: decline when the indexed source material can't support one.

Solution Design

We built three deployable layers with a clean contract: a Flutter client that never reaches the retrieval layer directly, a NestJS product API (BFF) holding accounts and progress in MongoDB, and a Python FastAPI RAG gateway with per-subject SQLite (FTS + embeddings) for exam-content retrieval.

Every generated answer, mark, or quiz question is grounded in retrieved source material - the API errors rather than inventing when the corpus can't support a response. A student photographs or types a past-paper question and gets a structured, examiner-style model answer with a marks breakdown and cited sources; submitting their own answer runs the same retrieval and marking pipeline, scoring awarded vs. total with strengths and missed points; and a live quiz checker builds topical or mixed sets on demand, scoring MCQs live and short answers through the same marking pipeline, updating spaced-repetition-style mastery as the student goes.

Implementation Process

01

Phase 1: RAG Gateway

Built per-subject SQLite retrieval (FTS + embeddings) behind a FastAPI gateway, with grounding enforced at every hop.

02

Phase 2: Product API & BFF

Shipped the NestJS product API holding accounts and progress, proxying all exam-content requests so the client never talks to retrieval directly.

03

Phase 3: Flutter Client & Delivery

Delivered the Flutter app for Android and iOS, wired Stripe Checkout for Pro/Premium, and shipped via AWS ECR to dev/QA deploy channels.

Results & Business Outcomes

Grounded, Not Generic

Every answer, mark, and quiz question traces back to indexed source material across 15+ Cambridge O & A Level syllabuses - the system declines rather than invents when it can't.

Examiner-Style Feedback, Instantly

Students get a marks breakdown with strengths and missed points on their own answers, plus a live quiz checker that scores as they go rather than only at the end.

Technology Stack

FlutterNestJSFastAPIMongoDBSQLite (FTS + embeddings)OpenAI / Azure OpenAIStripe CheckoutAWS S3Socket.IO

Learner Journey Flow

01
Answer a Question

Photograph or type a real past-paper part.

02
Retrieve & Author

Per-subject RAG grounds a structured model answer.

03
Check My Answer

Marked like an examiner - awarded vs. total.

04
Live Quiz Checker

Topical or mixed sets, checked as you answer.

05
Mastery Tracking

Weak topics resurface for spaced review.

ARCHITECTURE FIRST

Need an AI Strategy Before You Build?

Book a complimentary AI Architecture Review. We'll assess your use case, identify opportunities, evaluate technical feasibility, and provide clear implementation recommendations.