លំហសិក្សាកម្ពុជា
V1.0
លំហសិក្សាកម្ពុជា
V1.0
ReanMate is an AI-powered learning platform for people in Cambodia. The intended Minimum Viable Product (MVP) allows teachers to set up private classes, invite students to join online via a join code, upload PDFs and videos to the class, create assignments and quizzes, and manage each class's student usage of AI assistance. Students are able to enroll in authorized courses, view lessons created by the teacher, complete assignments and quizzes, and use AI tools to ask questions or get answers from lessons, quizzes, and assignments — all grounded in the teacher's uploaded content. The system is engineered to handle Khmer, English, or mixed-language questions, and to indicate the source document and page from which each AI answer is generated. This project illustrates advanced object-oriented programming concepts, including classes, service classes, repository classes, validation, composition, and separation of responsibility. It shows the structure of a layered system that decouples routes/controllers, business services, repositories, models, templates, and database access. Users, classes, memberships, lessons, videos, documents, assignments, conversations, messages, sources, practice questions, and AI usage records are stored in a relational database. The AI workflow involves processing and chunking PDFs, retrieving relevant embeddings, and building a grounded prompt that asks the OpenAI API or Gemini for an answer.
UNIVERSITY OF TECHNOLOGY & ENTREPRENEURSHIP
Faculty of Technology — BSc (Hons) in Computer Science
ReanMate
AI Tutor and Private Classroom
Report
Field | Detail |
Student Names | Yin Keo Odom, Thai Sokchea, Toek Pheakdey, Pou Chhiangheng |
Lecturer / Professor | Prof. Sek Sokcheat |
Academic Year – Term | Year 2, Term 5 |
Status | MVP in active development — live build available |
ReanMate is an AI-powered learning platform for people in Cambodia. The intended Minimum Viable Product (MVP) allows teachers to set up private classes, invite students to join online via a join code, upload PDFs and videos to the class, create assignments and quizzes, and manage each class's student usage of AI assistance. Students are able to enroll in authorized courses, view lessons created by the teacher, complete assignments and quizzes, and use AI tools to ask questions or get answers from lessons, quizzes, and assignments — all grounded in the teacher's uploaded content. The system is engineered to handle Khmer, English, or mixed-language questions, and to indicate the source document and page from which each AI answer is generated.
This project illustrates advanced object-oriented programming concepts, including classes, service classes, repository classes, validation, composition, and separation of responsibility. It shows the structure of a layered system that decouples routes/controllers, business services, repositories, models, templates, and database access. Users, classes, memberships, lessons, videos, documents, assignments, conversations, messages, sources, practice questions, and AI usage records are stored in a relational database. The AI workflow involves processing and chunking PDFs, retrieving relevant embeddings, and building a grounded prompt that asks the OpenAI API or Gemini for an answer.
Live Application: https://z-rean-mate-abum.vercel.app/
Source Code (GitHub): https://github.com/The-Prompt-Guys/zReanMate
The links above point to the team's working deployment and source repository for ReanMate, referenced throughout this report and shown in the UI Showcase section.
Digital learning systems increasingly combine course materials, assignments, communications, and personalized learning support. ReanMate implements a focused learning environment in which a teacher's own lesson material becomes the main knowledge source for the AI tutor, instead of treating AI as a general-purpose chatbot. The platform connects the AI experience directly to a private class and the teacher's uploaded PDF lessons.
● Students need additional explanations after class.
● PDF materials alone do not provide interactive learning support.
● General AI may provide answers outside the teacher's materials.
● Students need answers they can verify against the original lesson.
● Teachers need control over AI usage in their classes.
● Private class materials must be protected from unauthorized access.
● Khmer/English bilingual AI support is needed.
● Teachers need classes, lessons, PDFs, and assignments in one platform.
● AI usage must be controlled and monitored.
ReanMate will provide private classes, teacher-managed PDF lessons, assignments, and a controlled AI tutor that answers questions based only on authorized class materials.
● Implement secure login and role-based access.
● Manage lessons and PDF learning materials.
● Manage assignments and student progress.
● Build a PDF-grounded AI Tutor.
● Support Khmer and English learning.
● Provide AI answers with source-page references.
● Generate practice questions.
● Track and control AI usage.
● Apply secure OOP-based system architecture.
● How can object-oriented design be used to separate learning-domain responsibilities from database and web concerns?
● How can private class membership be enforced throughout normal requests and AI retrieval?
● How can teacher-uploaded PDF material be transformed into searchable chunks while preserving page information?
● How can a RAG workflow provide useful AI explanations while reducing unsupported answers?
● How can the system support Khmer, English, and mixed-language questions?
● How can AI usage be measured and limited during a small demonstration?
● Student, teacher, and administrator roles.
● Private classes with join codes.
● Class membership management.
● Teacher-created lessons and private PDF materials.
● PDF viewing and page navigation.
● Teacher-created and published assignments.
● Student assignment dashboard and status filtering.
● Class-scoped AI Tutor and Study Workspace.
● AI quick actions: Explain in Khmer, Summarize, Generate Practice Questions, and Quiz Me.
● Source document/page citation for grounded AI answers.
● Practice sets and questions.
● AI usage tracking and limits.
● Administrative monitoring of users, classes, documents, and AI usage.
● Live video classes.
● Native mobile application.
● Payments and subscriptions.
● School-wide administration.
● Public course marketplace.
● AI voice tutor.
● Advanced proctoring.
● Full LMS grading engine.
● Automatic high-stakes grading.
● Additional AI providers before the OpenAI-based MVP is reliable.
Role | Main Responsibilities |
Student | Join private classes, read lessons, view assignments, ask the AI Tutor, practice, and monitor learning activity. |
Teacher | Create classes, manage members, upload lessons, create assignments, and configure class AI behavior. |
Admin | Manage users and classes, monitor document processing and AI usage, apply system controls, and review technical logs. |
Lecturer / Evaluator | Review the R&D process, OOP design, implementation, testing, and final demonstration. |
The project uses object-oriented programming to represent important concepts such as User, Class, Lesson, Document, Assignment, Conversation, and AI Usage. The goal is not to create classes only for appearance — each class has a clear responsibility and meaningful methods. Services coordinate business workflows, while repositories isolate database operations.
Composition is preferred where one component depends on another, such as an AI service depending on a repository or usage tracker. Inheritance is used only when the relationship is meaningful, such as role-specific behavior, where it improves the design. Type hints, validation, and documentation are used for public methods.
Python is the main programming language for the course implementation. Its class system, type hints, modules, exception handling, testing ecosystem, and web frameworks make it suitable for demonstrating object-oriented design.
Flask is used as the web framework required by the Advanced OOP course. Flask routes/controllers receive requests, validate access, call service methods, and return HTML views or JSON responses. The route layer does not contain direct SQL queries or complex business rules.
PostgreSQL is used as the primary relational database for the course implementation. Relational tables represent users, classes, memberships, lessons, documents, assignments, conversations, messages, and AI usage. Primary keys, foreign keys, unique constraints, and appropriate data types are used to maintain data integrity.
ReanMate uses a retrieval-augmented generation (RAG) concept. Teacher PDFs are processed page by page, cleaned, divided into chunks, and represented for semantic retrieval. When a student asks a question, the application checks authentication, class membership, AI availability, and usage limits before retrieving relevant material. The retrieved context is then used to construct a grounded prompt for the AI provider.
A key requirement is that retrieval must remain inside the student's authorized class/document scope. The AI system must not retrieve chunks from another private class. When the material does not sufficiently answer a question, the tutor states that the material is insufficient rather than inventing a source or page.
The product specification selects the OpenAI API as the MVP AI provider. The API key remains on the backend and is never exposed to the browser. AI requests, token usage, and estimated cost are recorded for monitoring.
● Register
● Login
● Logout
● Password management
● Student role
● Teacher role
● Admin role
● Role-based access control
● User list
● CRUD
● Search
● User report
● Class list
● CRUD
● Search
● Class report
● Lesson list
● CRUD
● PDF upload/delete
● Search
● Processing status
● Assignment list
● CRUD
● Search/filter
● Assignment report
● Ask AI
● PDF-based retrieval
● Source/page citation
● AI usage tracking
● Practice list
● CRUD
● Search
● Practice results
● User management
● Class monitoring
● AI usage monitoring
● System controls
Figure 1. Educational Platform System — module breakdown.
Class | Responsibility | Example Methods |
User | Represent account identity, role, and status. | has_role(), is_active() |
Classroom | Represent a private learning class. | generate_join_code(), archive() |
ClassMember | Represent membership between a user and class. | can_access() |
Lesson | Represent an ordered published/unpublished lesson. | publish(), unpublish() |
Document | Represent an uploaded lesson PDF and processing state. | validate(), mark_ready() |
DocumentChunk | Represent page-aware searchable document content. | metadata() |
Assignment | Represent teacher-created class work. | publish(), close(), is_overdue() |
AssignmentProgress | Track a student's assignment state. | complete(), submit() |
Conversation | Represent a student's AI conversation. | add_message() |
Message | Represent an AI/student message. | record_usage() |
MessageSource | Connect an AI message to a document page/chunk. | open_reference() |
PracticeSet | Represent saved practice generated from material. | add_question() |
PracticeQuestion | Represent one practice question. | check_answer() |
AIUsage | Record request type, model, tokens, and estimated cost. | within_limit() |
Service | Responsibility |
Auth Service | Registration, login, password verification, session-related business rules. |
Class Service | Create classes, join codes, membership, archive and access rules. |
Lesson Service | Create/publish lessons and coordinate lesson documents. |
Document Service | Validate uploads, coordinate private storage and processing status. |
Assignment Service | Create/publish/close assignments and manage assignment state. |
AI Service | Coordinate AI requests, prompts, answers, citations, and usage recording. |
RAG Service | Retrieve authorized document chunks and prepare grounded context. |
Practice Service | Generate and store practice sets and questions. |
Usage Service | Apply request limits and record token/cost information. |
Repositories provide database access without UI logic. Example repositories include UserRepository, ClassRepository, ClassMemberRepository, LessonRepository, DocumentRepository, AssignmentRepository, ConversationRepository, PracticeRepository, and AIUsageRepository. Parameterized SQL or a safe database abstraction is used instead of concatenating user input into SQL.
project_name/
├── app/
│ ├── __init__.py # Flask application factory
│ ├── config.py # configuration classes
│ ├── extensions.py # db, login_manager, bcrypt, etc.
│ ├── models/ # domain/data models
│ ├── repositories/ # MySQL data access
│ ├── services/ # business rules / workflows
│ ├── routes/ # Flask blueprints
│ ├── forms/ # WTForms or validation helpers
│ ├── templates/ # Jinja2 pages
│ └── static/ # CSS, JS, images
├── tests/
├── migrations/ or sql/
├── docs/
├── run.py
├── requirements.txt
└── README.md
The request flow is Browser → Flask Routes/Controllers → Services → Repositories → PostgreSQL. The AI flow adds document processing and retrieval components: PDF Upload → Validation → Private Storage → Page Extraction → Chunking → Embedding/Indexing → Authorized Retrieval → Grounded Prompt → OpenAI API → Saved Answer + Sources.
The API owns every data access decision. PostgreSQL is the durable state store; Redis and MinIO solve separate operational problems and never hold the authoritative learning record.
● DATABASE_URL and storage credentials stay in the backend environment only.
● File bytes are not database columns — relational rows keep private object metadata.
● PostgreSQL remains authoritative even when Redis or MinIO is temporarily unavailable.
● One teacher can own many classes.
● A class can contain many students through ClassMember.
● A class can contain many lessons.
● A lesson can have one or more associated documents depending on final implementation.
● A document can contain many page-aware document chunks.
● A class can contain many assignments.
● An assignment can have progress records for many students.
● A conversation belongs to a user and can be scoped to a class/lesson/document.
● A conversation contains many messages.
● An AI message can have multiple source references.
● AIUsage records connect usage to a user and, optionally, a class.
User Interface → Student Experience → Teacher Experience → Admin Experience → AI Processing Workflow
Teacher PDF upload workflow and student question workflow.
● Teacher material is the primary evidence.
● If the material is insufficient, the system clearly says so.
● The AI must never fabricate a source page.
● Explanations should be readable and appropriate for students.
● Khmer, English, and mixed-language questions are supported.
● Useful English technical terms may be preserved in Khmer explanations.
● For assessed assignments, the tutor should prefer explanations, hints, examples, and step-by-step guidance rather than automatically providing final answers.
Area | Example Endpoint |
Authentication | POST /auth/register, POST /auth/login |
Classes | GET/POST /classes, POST /classes/join |
Students | GET /classes/<id>/students |
Lessons | GET/POST /classes/<id>/lessons |
Documents | POST /lessons/<id>/documents |
Assignments | GET/POST /classes/<id>/assignments, PATCH /assignments/<id> |
AI | POST /ai/ask, POST /ai/practice |
Admin | GET /admin/ai-usage |
● Hash passwords securely.
● Keep the OpenAI API key on the backend only.
● Apply server-side role and ownership guards.
● Validate resource access on every protected request to prevent IDOR.
● Use private/signed document URLs where applicable.
● Validate PDF MIME type and maximum file size.
● Rate-limit authentication, upload, and AI endpoints.
● Log administrative actions and AI errors without logging secrets.
● Keep classes private by default.
Limit | Demo Setting |
Total demo users | 10 |
AI Tutor | 30 requests/user/day |
Burst rate | 3 AI requests/minute/user |
Monthly user ceiling | 500 AI requests/user |
AI lesson | 2 sessions/day if enabled |
Practice generation | 3 sets/day |
Test | Expected Result |
Teacher creates a class | Class is created and a join code is available. |
Student joins valid class | Student becomes a class member. |
Student uses disabled/archived code | Join is rejected. |
Teacher uploads valid PDF | PDF is accepted and processing begins. |
Teacher uploads invalid/oversized file | Upload is rejected with a clear error. |
Student opens published lesson | Authorized student can view the PDF. |
Teacher publishes assignment | Assignment appears to authorized students. |
Student asks grounded AI question | Answer is returned with a source document/page. |
Khmer/English/mixed question | Tutor responds appropriately in the requested language. |
Admin views AI usage | Usage and estimated cost are visible. |
● Student attempts to access another class by changing a URL ID.
● Student attempts to access another class's document directly.
● Teacher attempts to edit another teacher's class.
● User attempts to change their role through a frontend request.
● AI endpoint is called above its rate limit.
● Vector retrieval attempts to cross a class boundary.
● Non-PDF and oversized uploads are submitted.
● Question is directly answered in the PDF.
● Question is partially answered in the PDF.
● Question is absent from the PDF and the AI should state that the material is insufficient.
● Khmer-only question.
● English-only question.
● Mixed Khmer/English question.
● Student makes an incorrect assumption and the tutor corrects it using the source material.
● Source page opens the correct PDF location.
Phase | Main Work | Acceptance Evidence |
0 – Foundation | Repository, Flask frontend/backend setup, MySQL, environment, lint/test baseline. | Application and database run together. |
1 – Authentication | Register/login/logout, roles, protected routes. | Role-based access works. |
2 – Private Classes | Create class, join code, join/leave, student list, archive. | Class flow works without AI. |
3 – Lessons & PDFs | Lessons, PDF upload/storage, validation, viewer, processing status. | Student securely opens teacher PDF. |
4 – Assignments | Teacher create/publish; student dashboard and assignment page. | Deadlines and visibility work. |
5 – Document AI Processing | Page extraction, chunking, embeddings/indexing. | Relevant chunks are retrieved. |
6 – RAG + OpenAI | Grounded prompt, API call, citations, no-answer behavior. | AI answer includes correct source. |
7 – Study Workspace | PDF + AI split workspace, quick actions, language modes. | Approved learning UI works. |
8 – Practice | Generate/save practice questions and explanations. | Practice can be reused. |
9 – Usage Controls | Rate limits, token/cost tracking, admin controls. | AI usage remains controlled. |
10 – QA & Demo | Security, responsive testing, ten-user accounts, real lesson validation. | Reliable demonstration. |
Week | Task | Deliverable |
1 | Finalize scope, team roles, and requirements. | Confirmed project direction. |
2 | OOP class design, architecture, ERD, and Flask foundation. | Class diagram, ERD & project skeleton. |
3 | Authentication, roles, private classes, and membership. | Working login, access control & class module. |
4 | Lessons, PDF upload, and Assignments modules. | Lesson, PDF & Assignment modules. |
5 | PDF processing, chunk retrieval, RAG, and OpenAI integration. | Searchable document chunks & grounded AI Tutor. |
6 | Study Workspace development and UI refinement. | Student workspace. |
7 | Practice features and usage controls. | Practice tools & limits. |
8 | Security, functional testing, and bug fixing. | Test evidence & bug fixes. |
9 | Final bug fixing and demo preparation. | Stable MVP. |
10 | Final report writing and defense PPTX preparation. | Final submission. |
Risk | Impact | Mitigation |
AI API cost increases | High | Rate limits, daily/monthly caps, usage tracking, global AI control. |
AI gives unsupported information | High | RAG grounding, source citations, insufficient-material behavior. |
Cross-class data exposure | Critical | Membership guards, ownership checks, retrieval filters, security tests. |
PDF processing failure | Medium | File validation, processing status, error handling. |
Scope becomes too large | High | Keep MVP exclusions and prioritize core proof. |
Team member availability | Medium | Clear responsibilities and shared Git activity. |
Security mistakes | High | Server-side authorization and dedicated security testing. |
● Classes are private by default.
● Documents are not exposed via public URLs.
● Student conversations are not exposed to other students.
● Teacher analytics should prefer aggregate signals rather than reading private chats.
● Administrative access to user content should be minimal and audited.
● Technical logs should avoid secrets and unnecessary private content.
ReanMate is intended to support learning rather than replace the student's own work. For marked assignments, the tutor prefers explanations, hints, examples, and step-by-step guidance rather than automatically providing a final answer. The system also communicates uncertainty when the teacher's material does not contain enough information to answer.
The screenshots below are taken from the live ReanMate deployment (see Project Links) and show the mobile-first UI implemented so far for the student experience: the home dashboard, the learning-document library, joining a class, the AI study assistant chat, and the profile/settings screen.
Home Dashboard  Welcomes the signed-in student by name, offers a one-tap “add your material” shortcut for uploading a PDF, and lists the student's document folders (e.g. “cake”, “data”). | Learning Documents Library  Lists every document folder the student has created, each showing how many files it contains, with a floating action button to add a new folder or document. |
Classroom — Join a Class  The Classroom tab, where a student who has not yet joined a class is prompted to enter one using a join code, matching the private-class membership flow described in Chapter 2. | AI Study Assistant  The class-scoped AI Tutor chat. A document/class selector at the top constrains retrieval, and the assistant opens by asking what the student wants help with, grounded in their own learning material. |
Profile and Settings  Shows the student's activity summary (documents studied, streak days, goal progress) and account settings such as notifications, session timeout, appearance, and language (Khmer/English). |
ReanMate implements a focused AI-assisted learning platform that combines private classes, teacher-controlled PDF materials, assignments, and a document-grounded AI tutor. The design directly supports the Advanced OOP course requirements by separating domain objects, services, repositories, routes, and database access. It also provides a practical R&D problem: how to make AI assistance useful while maintaining private class boundaries, source traceability, and controlled usage.
The MVP is deliberately limited to approximately ten users so that the team can prioritize a reliable end-to-end proof. The core demonstration is complete when a teacher creates a private class, uploads a real lesson, creates an assignment, a student joins the class, studies the PDF, asks a useful question in Khmer or English, receives a grounded answer with the correct source page, and completes the learning flow without manual database intervention.
A working build of the student experience is already deployed and browsable at the Live Application link in the Project Links section, with the full source available in the GitHub repository, and is illustrated in the UI Showcase (Chapter 3).
● Flashcards and improved AI lesson mode.
● Richer assignment submissions and a stronger progress page.
● Announcements and due-date notifications.
● DOCX/PPTX support.
● Mock exams and weak-topic detection.
● Teacher review workflow.
● School/organization accounts and institution analytics.
● AI voice questions, OCR, lecture transcription, personalized tutoring, study plans, and AI-generated diagrams.
The product plan recommends delaying payment systems, school administration, additional AI providers, and other broad LMS features until the core student and teacher value is proven.
[1] S. Sokcheat, “Advanced Object-Oriented Programming,” lecture notes, University of Technology and Entrepreneurship, Phnom Penh, Cambodia, 2026.
[2] GeeksforGeeks, “Object-Oriented Programming (OOPs) Concept in Java,” GeeksforGeeks, 2026. [Online]. Available: https://www.geeksforgeeks.org/object-oriented-programming-oops-concept-in-java/
មតិយោបល់ 0