We live in an era of information overload. We buy books we never open and save podcasts we never listen to. The intention is there—we want to learn—but the time simply isn't.
I noticed that people (myself included) often struggle to finish the content they consume. They might read the first chapter or listen to the first 10 minutes, but life gets in the way. I wanted to solve this by building SUMMARIZED: a platform that digests heavy content into accessible formats like video, audio, text, slides, and infographics.
Here is the story behind how I built it, the tech stack I used, and the surprising lessons I learned about working with AI.
The Tech Stack
To get this off the ground quickly without sacrificing performance, I stuck to a robust, modern stack:
- Framework: Next.js
- Styling: Tailwind CSS
- Backend & Storage: Supabase
While the web framework was standard, the real challenge was the content generation pipeline. How do you take a 300-page book and turn it into a slide deck or a podcast script automatically?
The Secret Weapon: NotebookLM
I realized I didn't need to build a custom LLM pipeline from scratch when Google’s NotebookLM existed. It became the engine of the operation.
My workflow looks like this:
- I gather the source material (books, articles, transcripts).
- I feed them into NotebookLM.
- I prompt it to generate summaries in specific formats: an audio overview, a video script, a slide deck outline, textual takeaways, and data for infographics.
- I export these assets and upload them to my database.
The Storage Constraint
One technical hurdle I hit early on was the Supabase free tier limit. Storing audio and images adds up fast. To stay within the free tier boundaries, I had to implement an aggressive compression strategy for all assets before they hit the storage bucket. It taught me to be efficient with data right from the start.
Accelerating UI/UX with AI
I am a developer, not a UI designer. In the past, I would have spent weeks agonizing over padding and color palettes. This time, I leveraged AI design tools like Lovable, Bubble.io, v0, and Stitch.
This was a game-changer. Instead of designing components pixel-by-pixel, I described what I needed, and the AI generated the UI code in minutes. This allowed me to focus entirely on my area of expertise: the logic and the architecture.
The "Google Antigravity" Framework
To keep the project grounded, I used Google Antigravity to help build the project based on a set of fundamental rules I had written.
If you've ever coded with AI, you know it can hallucinate or drift off-topic. By establishing strict "rules of gravity" (fundamentals) for the project, I ensured the AI assistants stayed aligned with the project's core architecture.
Key Lessons Learned
Building SUMMARIZED taught me two massive lessons about modern software development.
1. Rules Are Everything for AI
The most time-consuming part of this project wasn't coding—it was defining the rules.
I learned that if you want AI to build complex software, you cannot just say "build this." You need to provide a rigid framework of rules. Once I nailed down these rules, the AI's output became streamlined and predictable.
The best part? This setup time is an investment. Now that I have these rules defined, I can reuse this "rule-set" for every future project to get up and running instantly.
2. Component Libraries > Custom AI Components
I initially tried to get AI to build complex UI components from scratch. It often failed or produced buggy code.
I realized that Component Libraries are the solution. AI models are excellent at reading documentation. When I instructed the AI to use a specific library and provided the docs, the output was flawless. Documentation provides the context that trial-and-error simply can't.
Final Thoughts & Try It Out
SUMMARIZED started as a way to help people consume content, but building it became a masterclass in how to leverage AI for development. By combining Next.js with NotebookLM and a rule-based AI workflow, I was able to ship a complex product in a fraction of the time it would usually take.
Want to see the result in action?
Whether you want to inspect the build or simply catch up on your reading list without spending hours, check out the live application here:
Top comments (0)