codeMe
Feature engineering on codeMe, a Georgian e-learning platform — a Tiptap-based rich-content authoring library, AI-evaluated coding exercises, and the student experience end to end.
My Role
Full Stack Engineer — built the Angular rich-content library powering course creation, the AI-graded Coding Exercise feature, and the student interface and profile system across backend and frontend. Later engagements added a RAG pipeline (n8n, Pinecone, Gemini embeddings), a JWT/RSA-based SDK generation system for API key distribution, and dynamic per-course font loading with Georgian subset support.
Project Type
Product Engineering — e-learning platform features spanning authoring, assessment, and the student experience.
Tools Used
Angular, Tiptap, TypeScript, AI-assisted grading, n8n, Pinecone, Gemini embeddings, JWT/RSA, web font subsetting
Live Link
https://codemelive.com
Problem & Solutions
Rich Authoring for Non-Developers
Problem
Course creators needed magazine-grade lessons — rich text, embedded media, custom styling — but the platform offered only plain content fields, so every ambitious lesson became a developer ticket.
Solution
A Tiptap-based Angular rich-content library gave instructors structured editing with multimedia embedding and custom styling, turning lesson design into a self-service workflow.
Grading Code at Scale
Problem
Manual review of student code could not keep up with course growth, and slow feedback is where beginner momentum goes to die.
Solution
The Coding Exercise feature runs assignments in-platform and evaluates them with AI, returning instant feedback and grading the moment a student submits.
Retrieval Instead of Guesswork
Problem
Learners ask questions in the middle of lessons, and a generic chatbot answering from nothing produced confident but wrong replies that undermined trust in the platform.
Solution
A RAG pipeline built with n8n, Pinecone, and Gemini embeddings retrieves the relevant course material first, so answers are grounded in what the instructor actually wrote.
Architecture
The Angular platform with the Tiptap-based authoring library; AI grading for coding exercises, and an n8n RAG workflow over Pinecone and Gemini embeddings for student questions.
Decision log
The calls that shaped the build — what was decided, why, and what it cost.
Tiptap for lesson authoring
Decision. Build the rich-content library on Tiptap rather than a markdown or plain-text editor.
Context. Instructors needed magazine-grade lessons with embedded media and custom styling.
Consequence. Structured, extensible content with custom nodes; more engineering than markdown, paid back by every lesson authored without a developer.
AI-graded coding exercises
Decision. Evaluate submissions in-platform with an LLM and return feedback immediately.
Context. Manual review could not keep up with course growth, and slow feedback kills beginner momentum.
Consequence. Instant feedback and grading at scale; grading needs rubrics and cost control, reviewed as the feature evolved.
Retrieval before generation
Decision. Answer student questions through a RAG pipeline orchestrated in n8n over Pinecone and Gemini embeddings.
Context. An ungrounded chatbot produced confident wrong answers and undermined trust.
Consequence. Answers are grounded in instructor material, and the pipeline is editable in n8n without redeploying the platform.