How I leveraged AI tech to build Kala.Social
In this post, I describe how AI tech has been invaluable in building Kala's prototype and how quickly it all came together.
If you haven't already, please read my first post on “Why I’m building Kala.Social.”
The last decade of my life was spent heads-down in product work while letting my uber-talented engineering teams manage the vagaries of writing production code.
When it came time to build Kala, I had a choice - hire a developer team to quickly build an MVP or try my hand at software engineering again. I admit it seemed daunting, but after seeing demos of code co-pilots and cheap no-code platforms, I got inspired to at least try my hand at it. Plus, I'm familiar with consulting engagements, and finding a build partner who matches both your pace and quality standards is challenging.
Building Kala.Social
To accurately estimate the opportunity's size, I wanted to understand the marketplace as it exists today. As I mentioned in my last post, Indian traditional arts exist as a fragmented ecosystem, with events scattered across Facebook, websites, event platforms, WhatsApp groups, and email communities. Here are all the different ways I used AI tools to build the prototype.
I used OpenAI and Claude’s code generators to build a structured data parser in Python that operated on crawled HTML pages. This code inspected the page and extracted all the structured info from events like Title, Location, Artists, Date and time, and other details.
Code generators also helped me load all this into a Postgressql database for easy querying and searching.
Next, it was time to build a backend service that could serve all these events and allow users to browse these listings and take action. I knew Rails could be an option, but until a few months back, I had not written a single line of Ruby code, let alone any production code with it.
Github Copilot helped me get started on a basic Rails app, and I very quickly had a shell of an app running locally doing what I wanted. The next month, I had a bare-bones site with single sign-on, user controls, event listings, and the ability to manually post events.
It was time to deploy the app. I chose Heroku (because I’d heard so much about it) and very quickly encountered limitations like the read-only filesystem, missing build packs, and version mismatches between production and local environments. This was my first hurdle.
Thankfully, other engineers have documented their woes on platforms like StackOverflow and the RoR community. I was able to overcome these hurdles and have a server running on the cloud. Code generators struggle with deployment tasks and queries, and experimenting in production environments isn't feasible.
While prioritizing which events to aggregate, I noticed many folks create Flyers to advertise their events. I realized that ingesting flyers and pulling structured data from them could be a great way to scale and add new events to the site. I got to work on my first AI feature.
Here’s a quick demo:
Next, I used AI to generate structured Meta tags, SEO titles, and descriptions for these extracted events. I’m already starting to see early search traffic for very niche queries.
Building for Creators and Organizers
My next challenge is to build for the demand that I’m seeing, from all sides of the marketplace - consumers, creators & organizers. More on this in the next post. I’m convinced that this space needs tooling and education.
Tooling to help creators connect with their audience, tools to help organizers find and curate an audience, host events, manage ticketing, fundraising, payments, outreach and much more. Five out of nine events in the SF Bay Area are being managed through Google Forms and direct emails to organizers 😳
In conclusion, I want to acknowledge how gratifying it’s been building again. I’ve noted previously that AI tech will uniquely empower a particular class of builders & entrepreneurs; I’ve first-hand experience now. While it may seem trivial to build using AI, real businesses are still grounded in reality, and I can’t wait to see how Kala scales. I’ll leave you with this pencil sketch of Kala’s architecture for now. If someone had told me that building this seemingly simple experience would require eight separate services and a monolith, I might have been too intimidated to begin. However, Kala’s out and working great, ready to scale. Wish me luck!
