Motivation
I wanted to explore AI-powered content generation in a practical context: helping people extract structured knowledge from books. The idea was simple: upload a text, and the platform generates summaries, key concepts, and visual mind maps automatically.
What It Does
ConceptHub is a full-stack web platform where users can:
- Generate AI summaries from book texts using Google's Gemini API
- Create interactive mind maps that visually organize key concepts, themes, and relationships
- Save and share quotes with annotations
- Build a knowledge base with persistent storage and search
The platform handles the entire flow: text input, AI processing, structured output, and collaborative sharing.
Architecture
The technical stack reflects the choices I made to balance development speed with production quality:
- Frontend: React + TypeScript for type-safe, component-driven UI development
- Backend: Python service handling Gemini API integration and text processing
- Database: PostgreSQL on Vercel with SQL queries for data persistence
- Infrastructure: Docker containers deployed on GCP, with the frontend on Vercel
User authentication, session management, and content persistence are fully implemented. This isn't a demo, it's a functional platform.
What I Learned
This was my first full-stack project integrating a production LLM API. The main lessons:
- Prompt engineering matters more than model choice for structured output generation. Getting Gemini to produce consistent mind map structures required careful few-shot prompting.
- SQL over NoSQL for structured content: mind maps, quotes, and annotations have clear relational structure. PostgreSQL was the right choice over a document store.
- Deployment complexity scales non-linearly: Docker + GCP + Vercel worked well, but the operational overhead of managing multiple services was significantly higher than expected for a project this size.