Case Study | Building an AI exam preparation platform with personalised study plans and adaptive assessment

AI-Powered Exam Prep with Personalized Plans and Adaptive Tests

Generative AI Data Structuring Backend & API Development

FiftyFive built the AI backend for TutorBuddy, an exam preparation platform for CBSE students in grades 6 to 12. The platform diagnoses each student's knowledge, builds a day-by-day study plan around the exam date, generates study material and tests, and checks mastery before the student moves on. The engagement covered generative AI engineering, prompt design, backend API development and curriculum data structuring.

Build Your Platform With Us

The Client

Client overview

The client runs a tutoring platform for CBSE students in grades 6 to 12 and for competitive exam aspirants. Tutors invite students to specific subjects, and each student shares their syllabus and exam date during onboarding. The client wanted every student to receive a plan, study material and tests matched to their own level and time, without tutors preparing them by hand.

The Need

Challenges

Personalising exam preparation for every student meant generating accurate, curriculum-aligned content on demand while keeping plans realistic and assessments fair. Generic AI output would not follow the CBSE syllabus, a student's daily time limit or a consistent marking scheme.

Each student starts from a different level. The platform had to measure existing knowledge across the syllabus before planning, and identify specific weak topics rather than giving one overall score.
Study plans had to fit a fixed number of days before the exam and a daily study-hour limit. Every chapter had to be covered, with tests, revision and a final mock exam scheduled in the right order.
Weak topics needed more attention without leaving the rest of the syllabus uncovered, so plans had to prioritise and still be complete.
Study material had to suit the student's grade, the time allocated to each topic and the subject. Formula-based subjects such as Maths, Physics and Chemistry needed accurate formulas that the frontend could render.
Tests had to come in objective, subjective and mixed formats, with consistent evaluation and clear pass rules, so students could not move on before mastering a module.
Language model output had to be returned as strict, structured data that the application could use directly, with no free text that would break the interface.
A study assistant chatbot for students had to stay within the syllabus and refuse unsafe or off-topic requests.

Our Approach

Solution

FiftyFive built the client's backend as a set of FastAPI services that cover the full preparation cycle: diagnose, plan, learn, test, revise and assess. Chapter and topic data for every subject from grades 6 to 12 is stored in MongoDB and drives every generation request. OpenAI GPT-4o and GPT-3.5 Turbo generate questions, plans and content through prompts that return strict JSON for the frontend.

Curriculum data foundation

FiftyFive structured the curriculum so every AI output is tied to the real syllabus.

Chapter and topic data extracted for all subjects from grades 6 to 12.
Storage in MongoDB for fast retrieval by grade, subject and chapter.
One dataset driving study timelines, content generation and question papers.

Diagnostic testing and evaluation

FiftyFive built the onboarding assessment that sets each student's starting point.

An initial questionnaire generated from the student's syllabus.
Automatic evaluation that returns a score, identifies weak topics and sets pass or fail at a 60% threshold.
Weak-topic output passed directly into study planning.

Personalised study timelines

FiftyFive developed the planning engine that turns a syllabus and an exam date into a day-by-day schedule.

Two plan types: full-syllabus coverage, and a plan that puts weak topics first and then covers the rest.
Learning days with topics and time per topic, kept within the student's daily study-hour limit.
A test day after each chapter, followed by revision days once the syllabus is complete.
A three-hour full-syllabus mock test on the final day.

AI content generation

FiftyFive built content generation that adapts study material to the student and the time available.

Topic explanations sized to the hours allocated and written for the student's grade.
Formulas returned in LaTeX notation with explanations of each variable, ready for frontend rendering.
Worked examples, practice questions and reference links for every topic.
Separate revision content with key points, essential formulas and revision techniques.

Assessment and mastery checks

FiftyFive built the testing layer that confirms mastery before a student moves on.

Module question papers in objective, subjective or mixed formats.
Module evaluation with an 80% pass mark for topic mastery.
A three-hour, 80-mark final assignment that simulates real exam conditions.

Study assistant chatbot

FiftyFive added a chatbot that answers student questions about the material they are studying.

Answers grounded in the current study content, with conversation history for follow-up questions.
Off-syllabus questions redirected back to the topic.
Safety rules that decline harmful, explicit or inappropriate requests.

Technology

Technologies

Language & API

PythonFastAPI

AI / LLM

OpenAI GPT-4oGPT-3.5 Turbo

Database

MongoDB

Data format

Structured JSON outputsLaTeX formula notation

Deployment

Docker

Experts

Team

AI
Engineers

The Impact

Results

Project DurationOct 2024 – Nov 2024

FiftyFive delivered an AI backend that takes each student from a diagnostic test to a final mock exam with a plan, material and tests built for them. Tutors no longer prepare individual plans and papers by hand, and students follow a schedule that fits their exam date and daily time.

Personalised preparation for every student

Every student on the platform, from grade 6 to 12, gets a study plan built from their own diagnostic results.

Complete, realistic study plans

Day-by-day timelines cover the full syllabus within each student's daily hours, with chapter tests, revision and a final mock exam.

Mastery before progression

An 80% module pass mark and a three-hour final assignment confirm readiness before the exam.

End-to-end automation

Eleven API endpoints automate diagnosis, planning, content, testing, evaluation and student support.

Support

FAQs

What does TutorBuddy personalise for each student?+

The AI backend diagnoses knowledge gaps, builds a plan around the student's syllabus, exam date and available study time, then generates material, tests and revision content matched to that plan.

Who can use the platform?+

The client serves CBSE students in grades 6 to 12 and competitive exam aspirants. Tutors invite students to subjects, and students share their syllabus and exam date during onboarding.

How is a student’s starting level measured?+

An AI-generated questionnaire based on the student's syllabus is evaluated automatically. The assessment returns a score, identifies weak topics and uses a 60% pass threshold.

How are weak topics handled in study plans?+

The planning engine can put weak topics first and then cover the rest of the syllabus. It also offers a full-syllabus plan, so prioritising weak areas does not omit chapters.

How does the plan fit an exam date and daily schedule?+

The backend allocates learning days and time per topic within the student's daily study-hour limit. It schedules a test after each chapter, revision after syllabus coverage and a three-hour mock test on the final day.

What study material does the AI create?+

It generates grade-appropriate topic explanations sized to the allotted time, worked examples, practice questions and reference links. Revision content highlights key points, formulas and techniques.

How are Maths and science formulas returned?+

For formula-based subjects, the backend returns formulas in LaTeX notation with explanations of each variable, ready for the frontend to render.

What test formats are available?+

The backend creates objective, subjective and mixed module question papers. Module evaluation uses an 80% pass mark for topic mastery.

What is the final assessment?+

A three-hour, 80-mark final assignment simulates exam conditions. The timetable also includes a three-hour full-syllabus mock test on the final day.

How does the student assistant stay on topic?+

The chatbot grounds replies in current study content and carries conversation history across follow-up questions. It redirects off-syllabus queries and declines harmful or inappropriate requests.

How does the frontend receive AI output?+

GPT-4o and GPT-3.5 Turbo generate plans, questions and study content through prompts that return strict structured JSON, including LaTeX for formulas where needed.

What team and technologies delivered the backend?+

Two AI Engineers delivered the backend between October and November 2024 using Python, FastAPI, MongoDB, OpenAI GPT-4o and GPT-3.5 Turbo, with Docker for deployment.