readmap.ai

readmap.ai dashboard
example of generated book roadmaps
readmap.ai flowchart
generated roadmap as pdf

Inspiration

Readmap.ai began as a team-driven response to a shared challenge we faced as students and lifelong learners: figuring out what to read and in what order when exploring a new topic. We often found ourselves lost in a sea of book lists, blog posts, and forums, unsure where to start or which source to trust. We wanted something smarter, a system that doesn't just recommend books but builds a guided, personalized learning roadmap tailored to our goals and interests. This sparked the idea of building readmap.ai, a platform that could intelligently generate personalized reading roadmaps powered by AI.

What it does

Readmap.ai takes a user’s input, whether it’s a topic, a question, or a goal, and returns a structured reading roadmap, complete with personalized book recommendations arranged in a logical learning sequence. It adapts to the user’s interests, background, and depth of inquiry, making sure every roadmap feels custom-built. The interface is intuitive and visually appealing, providing an engaging experience through:

AI-generated roadmap logic
Interactive chatbot-like query input
Visually stacked book progressions
A clean and modern UI powered by Tailwind CSS

How we built it

Frontend: Next.js 13+ using the App Router and TypeScript, styled with Tailwind CSS and ShadCN UI for modern UI components.
Backend: Python setup scripts that initialize MongoDB Atlas and perform vectorization of books.
AI/Embeddings: Google Cloud’s Vertex AI powers the semantic understanding of user queries and book data.
Database: MongoDB Atlas stores and serves the book collection and vectorized embeddings.
Deployment: Hosted on Google Cloud Run for scalability and serverless deployment. The book data is pre-processed and vectorized using embedding models, then stored in MongoDB. When a user submits a query, a similarity search retrieves the most relevant books and arranges them based on topic progression.

Challenges we ran into

Finding a proper dataset: We had to curate and structure book data suitable for semantic search and learning progression.
Determining a satisfiable data structure: Mapping books to knowledge levels and themes required careful schema design.
Designing the collaboration between RAG (Retrieval-Augmented Generation) and vector query search: Merging semantic relevance with roadmap logic was tricky.
Front-end clashes: Building a functional UI that still looked polished within a tight timeline meant constant iteration.
Balancing visual appeal and usability: Making the drag-and-drop interactions, dropdown menus, and modals work seamlessly required cross-functional debugging.
Integrating the AI + database + UI stack: Each layer had different expectations and tooling; bringing it all together smoothly was a technical feat.

Accomplishments that we're proud of

Built a fully functional end-to-end product in a short time with modern frameworks and technologies
Developed a personalized AI assistant that feels intuitive and practical
Integrated Google Vertex AI for real semantic search, not just keyword matching
Implemented drag-and-drop reordering, renaming, and deletion of maps — all with responsive UI feedback
Achieved a polished UI with animations, gradients, and structured layout using Tailwind

What we learned

How to implement semantic search using embedding models and vector databases
The importance of design systems (like ShadCN and Tailwind) for rapid prototyping
Data modeling for AI systems that need progression logic, not just relevance
Real-world integration of cloud AI services, especially Vertex AI
How to collaborate effectively in a high-pressure environment while juggling frontend/backend/AI simultaneously

What's next for Readmap.ai

We plan to take Readmap.ai to the next level by:

Adding sign-in and authentication so users can save and sync their maps
Building saved maps persistence using cloud storage for cross-device support
Integrating a more dynamic AI assistant that adapts in real-time to user feedback
Expanding the dataset to include podcasts, videos, and articles beyond books
Adding a sharing feature so learners can collaborate and discover from one another