How I Created a Micro SaaS from Scratch in Just 7 Hours
A few weekends back, I built a simple tool for fun. Now, I have more than 30 customers trying and testing it. Here’s how it all started.
How It All Started
I have been using PowerPoint and Google Docs regularly since my academic days.
For the past 2–3 years now, I’ve been using Canva quite a lot.
With the integration of AI, there’s a tremendous opportunity to create slides in a short amount of time. Yes, most of the aforementioned tools have already integrated this feature to some extent.
But what if we make it even simpler and goal-oriented ?
What if we could create slides via a prompt with minimal features? Keeping all the flashy features out of sight and concentrating on the core function.
Idea : Prompt → Slide deck Generator
That’s it. No need to worry about design or background — all done by AI.Just prompt and get your decks ready in a minute or two. Next, you have the slide decks ready to share.
That’s how FastDeck AI started.
However, it’s now more important to validate the idea and just ship it to a couple of early users.
It worth to mention that I was very much inspired by Marc Lou and his approach to SaaS development.
“ … Instead, think in weeks, and ditch ideas that take longer than a month. You can also lower your time-to-market:
- Presell a product
- Ditch code testing & dark mode
- Break big ideas into 1-feature-only products
There is no perfect idea.The only way to know is to launch it.
This framework helps you increase the odds of success and get profitable faster… ”
Building FastDeck AI
I needed to use a tech stack that I’m well-versed in to build and prototype this product quickly. I chose Databutton.
Disclaimer: I work for Databutton and have been using this tool extensively over the last couple of months.
The Databutton AI app developer specialises in building full-stack web apps and MVPs conversationally, especially when the apps are AI-driven themselves.
With Python as the backend, React as the frontend, and one-button deployment, Databutton serves as a good starting point to build your application.
A Quick Sketch of the App Architecture
Spent the first couple of hours on sketching the main app architecture and setting up the key components:
- How the frontend would look
- Backend/API calls to make from the UI
- How the app would interact with the end-users
Also, I made sure to document each of the processes and share while I build.
Here’s the first tweet where I could make an AI Agent create a slide from scratch…
Iterating and Building the Minimal Features
Okay, we have something up!
The next step is to iterate further and get the minimal features app ready as early as possible.
I prioritzed next on the app’s UI more.
- Added some headers
- Polished the UI
- Made the “Generate Deck” button visible
- Changed the Background theme
- Added a “Logout” Button
Getting the Backend Work Done
With the app’s UI taking shape, it was time to focus on the Python backend and improve the main backend code to ensure smooth functionality and performance.
Implementing Core Functionalities:
- Firestore for User Authentication: Leveraged Firestore for managing user authentication. Databutton’s template was particularly helpful in setting this up efficiently.
- Agent Code Generation: Used a simple prompting hack with OpenAI’s GPT-4 to generate code faster. This approach worked remarkably well for speeding up development.
- Authentication + Firestore Database Setup: Spent time deciding on the best way to set up authentication and structure the Firestore database. Also, finding a scalable architecture that could grow with the app’s user base.
Key APIs (Backend) Necessary for the quick MVP
agent API: The core backend API responsible for creating slide decks based on the user prompt. Generates the slide decks and passes them to the frontend for display.
firestore_database API: Manages all user authentication and stores generated decks on a per-user basis.
Work in Progress: stripe_webhook API: Currently integrating payment functionality using Stripe. This will handle payment processing and subscription management
Polishing the MVP & Testing it Before Sharing with the World
This time, I focused on adding more low-hanging functionalities.
Features like the number of slides that can be generated or the tone of the slides (e.g., professional, casual) were introduced. These parameters are passed to the backend as part of the prompt.
Next, tested the MVP thoroughly to ensure all features worked as expected with the least number of bugs possible. I verified that the slide generation, user authentication, and other core functionalities operated smoothly.
Oh wait…
Also realized the need for a landing page to demonstrate all the possible features of FastDeck AI.
So, quickly generated an informative landing page that highlights the app’s key functionalities. Additionally, recorded a short video explaining how the app works.
More Promotion Over Social Media (X, LinkedIn, and TikTok) and Gathering Early Users’ Feedback
Only failed once, so I reduced the slides to 1; this is so much fun!!! and AMAZING!! nice work!
User feedback helped quite a bit! Based on the feedback, I added a new feature to share decks directly from the app.
In addition to social media promotion, I sent a few cold emails to early users.
How It Looks Now and What’s Next?
Now, we have more than 30 active users using FastDeck AI.
Somewhere along the way, I lost the momentum to promote it lately, but seeing the constant signups and positive feedback from the early users has been incredibly encouraging.
Yes, I would definitely love to work on FastDeck AI further. So, would love to hear your feedback and improve it!
Try it out and let me know what you think.
And most importantly, thank you for following along on this journey.
Stay tuned for more updates and improvements!
Here’s an update with Fastdeck AI and it’s open-source Python package —
https://medium.com/@avra42/building-an-ai-micro-saas-product-in-public-learnings-and-updates-80284ed24344
