We built an EdTech platform a complete learning management system with adaptive content delivery, a RAG-based AI tutor, automated assignment grading, and early dropout detection used by real students and teachers today.
An EdTech platform needed to serve four distinct user types students, teachers, parents, and institution admins each with completely different workflows and data access requirements, all in a single platform.
The web platform (React) is the primary surface for course delivery. Teachers can create courses with video lessons, PDF resources, text content, embedded quizzes, and assignment submissions. Students enrol, follow their course curriculum, submit assignments, take quizzes, and receive certificates on completion. The assessment engine supports MCQ, multiple-select, true/false, and short-answer question types with configurable scoring and pass marks. Certificates are generated from an HTML template and exported as PDFs automatically triggered when a student completes all required modules and passes the final assessment.
The mobile app (Flutter) extends the web platform for on-the-go learning. The key mobile-specific feature is offline download: students can download video lessons and PDFs for offline viewing, with progress synced back to the server when connectivity is restored. The mobile video player is built with custom controls playback speed adjustment, chapter markers, and continue-from-last-position. Push notification reminders for upcoming assignment deadlines and live session schedules are managed through Firebase Cloud Messaging.
The teacher dashboard provides per-student progress visibility down to the individual lesson level: which lessons watched, what score achieved on each quiz attempt, which assignments submitted vs. overdue. Teachers can grade open-ended assignments, post announcements to a cohort, manage attendance for live sessions, and send direct messages to individual students or parents. The parent portal shows a read-only view of their child's progress, upcoming assignments, attendance record, and teacher messages with weekly email summaries configurable by the parent.
Live sessions are managed through a scheduling interface: the teacher creates a session, the platform generates a Zoom or Google Meet link (via respective APIs), adds it to all enrolled students' calendars, and tracks attendance post-session by comparing join/leave times from the meeting platform's webhook data.
Six modules that cover every step of the learning journey from course creation to certificate issuance for students, teachers, parents, and admins.
Teachers build courses from video lessons, PDFs, text modules, embedded quizzes, and assignment submissions. Students follow a structured curriculum with progress tracking per lesson. Certificates auto-generated on completion from customisable templates.
MCQ, multiple-select, true/false, and short-answer question types with configurable scoring, time limits, and pass marks. Multiple attempts configurable per quiz. Detailed per-question analytics show which questions students are consistently getting wrong.
Flutter app with downloadable lessons for offline access. Custom video player with speed controls and chapter markers. Progress syncs automatically when connectivity returns. Push notification reminders for deadlines and live sessions via Firebase FCM.
Teachers schedule live classes; the platform creates Zoom or Google Meet links via API, adds sessions to enrolled students' calendars, sends reminders, and records attendance by processing join/leave webhook data from the meeting platform.
Teacher dashboard with per-student per-lesson progress, grading interface for open-ended assignments, attendance management, and announcement broadcasting. Parent portal with read-only access to their child's progress, attendance, and teacher messages.
School or institution-level management: user management (students, teachers, batches), course assignment to batches, revenue reporting, and platform usage analytics. Multi-institution support with data isolation for edtech platforms serving multiple schools.
These aren't experimental features they run in production and directly affect learning outcomes, teacher workload, and student retention.
After each quiz, the adaptive engine evaluates the student's score against thresholds: a student scoring above 85% is ready for accelerated content and is advanced to a harder module track. A student scoring below 60% is given a set of remedial practice exercises that target the specific weak areas identified from their wrong answers before they're permitted to proceed to the next module. The middle band gets the standard progression. This means two students enrolled in the same course can be on different content tracks based on demonstrated performance.
Result: improved course completion rate for all performance bands
Students type questions in natural language inside any lesson. The doubt bot retrieves the most relevant chunks from the course content using Pinecone vector search (embeddings generated via OpenAI), then passes the retrieved context plus the student's question to GPT-4 to generate a precise, grounded answer. Responses are sourced exclusively from enrolled course content the bot won't answer questions outside the scope of the course material. When a question cannot be answered from course content with sufficient confidence, it is escalated to the teacher with the student's question and the bot's confidence score as context.
Result: 67% of student doubts resolved without teacher intervention
MCQ and multiple-select assignments are graded instantly by the system with score calculation and feedback generation. Short-answer questions are graded using an NLP model that compares the student's answer against a reference answer set, checking for conceptual coverage rather than exact string matching. The model outputs a score and a short explanation of why marks were awarded or deducted, which the teacher can edit before releasing to the student. Teachers review AI-graded short answers in a batch review interface rather than grading from scratch.
Result: 55% reduction in teacher time spent on objective assessment grading
An ML model monitors three behavioural signals for each enrolled student: login frequency (has the student not logged in for 5+ days when they used to log in daily?), quiz performance trajectory (is their quiz score declining over the last 3 assessments?), and assignment submission rate (are they submitting less than 50% of assignments in the current week compared to their own historical average?). When two or more signals align, the student is flagged as at-risk in the teacher's dashboard with a "last seen" timestamp and their recent engagement metrics. The flag appears approximately 2 weeks before a typical dropout, giving teachers time to intervene.
Result: dropout flags appear 2 weeks before typical disengagement, enabling early intervention
Full LMS platforms with course creation, video delivery, assessments, certificates, and analytics for schools, coaching institutes, corporate training, and edtech startups.
Flutter apps with offline content downloads, adaptive video players, push notification engagement loops, and progress sync. Built for learning on mobile networks with intermittent connectivity.
RAG-based chatbots that answer student doubts from course content using vector search and LLMs. Responses are grounded in the actual course material not generic internet knowledge so answers are accurate to the specific syllabus being taught.
Test and exam engines with proctoring support, configurable question types, automated grading, percentile scoring, and certificate generation. Built for competitive exam prep platforms and professional certification bodies.
Per-student, per-cohort, and platform-wide analytics on engagement, assessment performance, and completion rates. Predictive models for at-risk student identification. Data dashboards for teachers, institution admins, and product managers.
Zoom and Google Meet integration for live class scheduling, automated link distribution, calendar invites, attendance tracking from meeting webhook data, and recording management for post-class replay.
We've shipped a full LMS with adaptive learning, a RAG-based AI tutor, automated grading, and dropout prediction. Whether you're starting fresh or adding AI to an existing platform, let's talk through what you need.