Building an AI Micro SaaS Product in Public— Learnings and Updates
I recently shared my journey of building Fastdeck AI. Today, I’ll answer the questions I received.
Here’s an outline on what I’ll share today:
Current state of Fastdeck AI
Did Fastdeck AI gain a lot of users? No.
Is Fastdeck AI growing? Yes.
What’s my take? Honestly, I’m satisfied …
I have been following the journeys and drawing inspiration from many Indie hackers like Marc Lou, Bhanu Teja, Sacha and many others.
And, one thing I’ve learned is that building and gaining organic users isn’t easy. Gaining paid customers is even tougher! But it’s a process that pays off over time, and you learn something amazing along the way.
Last week, I sat down with Sacha to learn about his experience in building a SaaS product from scratch and selling it for six figures within a year!
Feedbacks from the Last post
This is the best part, and I adore it the most.
First, thanks for all the kind comments and DMs.
Building in public has an unique feeling — zero revenue but immense satisfaction while one shares and learns with hundreds of others.
Few of the most common technical queries I received about the app …
In short, the fastdeckPython package does all the heavy lifting.
Yes, that’s what I’ve open-sourced last week …
… But first, let me address these excellent questions above!
What’s the “AI Agent” in Fastdeck AI
In the last blog, I shared the initial crude architecture, which was then fed to Databutton to build the full-stack app — a Python backend with a React frontend.
The app has a single core backend, which is the Agent.
Question 1:
What is this Agent API?
In simple terms, the agent comprises of the following:
- Large Language Model
- Tools — a) Python code execution tool b) HTML code execution tool
- Custom Prompts fine-tuned to tweak the outputs
The LLM decides which tool to use. The custom prompts under-the-hood prioritise Python code execution over HTML code generation.
Learning; gpt-4-omni works better with tool calling. That’s the LLM behind this slide deck code generation!
Question 2:
… It takes a while for your openAI API to give the output so won’t that cause performance bottlenecks in the application? …
Yes! That’s exactly why the fastdeckPython package was built. The Python code execution tool mentioned above can execute Fastdeck SDKs. Thus the AI , can generate slide decks with just a few lines of Python code, saving time, instead of generating vanilla HTML code.
The HTML code generation tool is used as a fallback when the other fails. When this tool call gets triggered, the generation time is longer!
Summer holidays & Open-sourcing Fastdeck Engine
Good weather is scarce here in Scandinavia. But when it’s summer, this is the best place to be.
Did I feel like working during these days? Yes!
Did I work? No!
Took a two-week break.
I made the right choice.
I had no clear idea of what to work on, even though my brain was pretty much pumped to tackle everything.
The next day after the summer break, first thing — I wrapped up the initial functions that I had been using in the Fastdeck AI app into a Python package. Tested Fastdeck AI app with this package.
And …
Fastdeck AI now runs on fastdeck Python package.
You can use it too.
Locally or over your Python notebooks. I’ve already added few examples (over the repo) to get started with, and generate slides in pure Python syntax.
pip install fastdeck ( more in the docs here. )
Why open-sourcing?
- Providing Value: I don’t find many open-source tools similar to this, so I wanted to fill that gap and help others
- Growth with Maturity: As Fastdeck AI App matures, this open-source library will grow side by side, benefiting everyone
- Ease of Usage and Fast shipping : Making easy for me to test and deploy Fastdeck AI quicker
- Cross Traffic: Open-sourcing will allow a proportion of cross-traffic to Fastdeck App
What’s next with Fastdeck AI?
So, here’s my to-do’s for the coming days …
Adding Pricing. Learnings: I’ve realised that offering a free plan was a mistake, and I want to fix it ASAP. I’m ditching the free plan soon! ( Why ? Read this amazing post here from Marc Lou)
Would love to hear your thoughts and gauge your interest. Please join the waitlist and share your feedback about Fastdeck AI. Join the Waitlist for early access!
Final Thoughts
Let’s see how Fastdeck AI grows. More than that, I’m bullish about documenting this journey.
Time is the limiting factor, but if it’s fun, why not?
Through these posts, I aim to keep the process of building a micro SaaS product transparent and, most importantly, share what I learn along the way. I learned a simple rule from this video:
Ship Fast → Iterate → Get 1% Better
I hope this inspires and encourages you to start building your own.
Stay tuned for more updates, and thank you again all for your support and feedback!
