Welcome back to another post of Tech Break by Friday, where we break down big tech topics in small, digestible moments. I am Paraskevi Kivroglou, an experienced software developer and AI engineer. Even though I have the experience, I joined the hackathon and I used only Lovable to test how easy it is for people to use it and vibe-code their idea in an app. If you haven’t listened to our episode about vibe-coding, this is the link.
My idea for today’s episode is to break down the steps for building your own AI project. The goal is to help people advance their skills and gain more confidence in themselves using AI tools.
AI Tool - Lovable
Before we discuss the challenges I faced building an AI-powered research assistant, I want to give a quick shoutout to a tool that made this whole process surprisingly fun and accessible: it’s called Lovable.
Lovable is a platform that lets you rapidly prototype and build full-stack AI applications just by prompting. It’s like having a co-developer who understands natural language, which is basically what you are telling it to do.
Now, just to be clear, this isn’t a sponsored segment. I’m mentioning Lovable because it genuinely helped me go from idea to working prototype on a weekend during a hackathon I joined.
I’ll leave a link in the show notes if you want to check it out and maybe build something cool yourself. I will also leave a link to my project if you're going to check it out.
Screenshot Lovable.dev Home Site
Problem & Inspiration
I am writing my Master’s Thesis, and Zotero is a tool for saving papers, articles, and any bibliography that you would like to mention or cite. As I was writing, I was reading something, but with too many papers open, I forgot where I had read this specific information. I then went through everything again until I found it, and I was losing too much time.
My idea was okay: What if I could make a fast tool that might not be perfect now? It works 80% of the time, which is good enough for me to use it.
So, I decided to join the Lovable Hackathon since it required a 48-hour submission. If I had time pressure, I would have actually built the project. I also posted about the everyday project on X (former Twitter).
What I Built – The MVP
1. User Authentication with Clerk
First, I integrated Clerk to handle user authentication. This meant that users could securely log in, and all their data — chat history, preferences, connected Zotero accounts — would be private and tied to their profile.
2. Supabase as the Backend
On the backend, I used Supabase — an open-source alternative to Firebase — to store user data, chat histories, and Zotero content. This gave me an instant, scalable database and real-time features right out of the box.
3. Zotero Integration
One of the core features was pulling in a user’s Zotero collections and papers. Once authenticated, the assistant would fetch the user’s library — including metadata like title, authors, abstract, and even tags — so it could be used in the chat context.
4. Secure API Key Management
I also built a secure way to manage API keys. Since this app uses Zotero’s API and Perplexity’s LLM interface, users can store their keys privately, and they are never exposed client-side. That was important for maintaining trust and data security.
5. Core Chat Interface
Now, the star of the show: the chat interface.
Users could interact with the assistant using natural language, things like:
Summarize this paper
Ask a question and generate an answer regarding the papers in the chosen collection (a collection here is a group of documents/links)
The AI would then respond intelligently by referencing actual papers from their Zotero library.
6. Persistent Chat History
Finally, I added persistent chat history. Each user could return to their previous questions and conversations. This meant you could ask a question in the morning, come back later, and continue the same train of thought, almost like having a virtual research assistant who remembers everything.
So that was the MVP: log in, connect Zotero, talk to your papers, and pick up where you left off.
How you can do it yourself
I built everything I mentioned before with natural language by asking the lovable chatbot. I will go through a series of things you can do by yourself.
You can chat with Lovable next to your project. By default, Lovable will usually output code or change the code.
Screenshot of how the chat and the project look together in the app
When you want to chat without making changes to your code, make sure you have chat enabled like this:
If you activate the toggle Code viewer, as shown below, you can view the actual code if you are interested.
Next, you can find a button to integrate Supabase, which is an open-source database that allows you to save data for your app.
5. If you want to monitor your application’s changes, you can allow Lovable to integrate your project in a GitHub repository.
The Invite Button allows you to invite your friends to collaborate.
7. The publish button allows your application to be live and shared with other people online.
This may sound difficult and confusing. In order to help you take action, I am sharing my first prompt of this project.
First prompt of Zotero Agent
I started with a simple idea: make it easier to talk to my academic papers. And what came out of it was a real MVP that’s already helping me speed up literature reviews, summarize research, and organize my thinking.
If you enjoyed this episode, please follow the podcast, leave a review, or share it with someone who’s been meaning to build something with AI but doesn’t know where to start. If you create something, please share it with me, either by dm or in the comments.
Thanks again for listening. I’ll see you next Friday, where we'll continue to explore new topics and ideas.
Sources:
Media:
Spotify | Instagram | Apple Podcasts | X