🎓 Moodle AI — Academic Mentoring Assistant

A RAG (Retrieval-Augmented Generation) pipeline embedded in a Moodle-integrated LMS that answers student queries using uploaded course materials, streams responses token-by-token via SSE, and delivers them through a Flutter mobile app.

• Smart Q&A Retrieval: Extracts and vectorizes multi-format course materials (PDF, DOCX) via NVIDIA NIM.
• Vision-Driven Parsing: Extracts tabular data and calendars using NVIDIA Vision LLM.
• Dynamic Query Routing: Intelligently routes queries between vector search, SQL calendar lookup, and LMS APIs.

• SSE Streaming: Delivers token-by-token stream responses with active source references.
• Mobile Chat App: Cross-platform Flutter application with history persistence and markdown formatting.
• Admin Dashboard: Simple web UI to upload documents, manage subjects, and configure models.

1. Student Query: Student asks a question via the Flutter application chat screen.
2. Intent Classification: Backend classifies query intent (concept_explanation, calendar_query, or student_data_query).
3. Retrieval Search: System executes pgvector similarity search, SQL calendar lookup, or LMS API retrieval.
4. LLM Generation: NVIDIA NIM LLM synthesizes an answer using the retrieved context.
5. Client SSE Stream: Flutter receives token-by-token server-sent events (SSE) and displays them in real-time.

Supabase acts as the primary data store and vector database, combining relational integrity with vector search capabilities:

• Course Material Vectors (pgvector): Stores chunked course text mapped to 1024-dimensional embeddings generated via NVIDIA NIM.
• Relational Tables: Manages data for academic subjects, document metadata, logs, and calendar event schedules.

Here is a preview of the student client application interface:

Deployed live on AWS EC2 with a full CI/CD pipeline (GitHub Webhook → Jenkins → Docker).

GitHub push on main triggers a webhook to Jenkins, which runs 6 stages: Checkout Code → Cleanup Docker → Build Docker Image → Remove Old Container → Run Docker Container → Health Check. Full pipeline completes in ~2 minutes.

Webhook — Last delivery successful

Jenkins Pipeline — 6-stage Docker CI/CD (latest run #11, all green)

Docker container live on EC2 — health check confirmed

$ docker ps
CONTAINER ID   IMAGE               COMMAND                STATUS             PORTS                                       NAMES
3e3d29e48392   moodle-ai-backend   "docker-entrypoint.s…" Up About a minute  0.0.0.0:3000->3000/tcp, [::]:3000->3000/tcp moodle-ai-container

$ curl http://43.206.238.148:3000/health
{"status":"ok"}

The backend runs as a Docker container (moodle-ai-container) built from server/Dockerfile, exposing port 3000. Managed via docker-compose and orchestrated by Jenkins on every push.