16 April 2025
🚀 Start your engines — 3:30 PM
After a classic countdown, Team Skyra (now skar or kars or raks or arks) huddled up to dive into our idea.
It started with a simple question: can we automate merchant outreach using AI? But the bigger challenge was figuring out how. That’s what we spent the next three and a half hours unpacking.
The B-plot: As we missed sleep (and trains) to get to Bangalore, we weren’t exactly at our sharpest. Sleepy heads and oncoming migraines hit us hard - thankfully, nothing a little Paracetamol and coffee couldn’t fix. As our hands fiddled with the fresh batch of stickers we’d just been gifted , our brains looped on the problem statement, again and again, until we finally cracked it.
The problem: Right now, our Key Account Managers (KAMs) and marketing folks slog through long Excel sheets, manually checking if merchants have enabled basic wishlist touchpoints. It’s tedious, repetitive, and definitely not scalable.
The final plan? Automate merchant outreach for wishlist adoption using AI.
The idea was to pull in site revenue, sessions, orders, and feature adoption data, and train an AI model to suggest the next best feature to enable — then automatically draft a personalized email for the outreach team to send.
Simple, right? Spoiler: it wasn’t.
🚧 Off we go — 7:00 PM
We split into two squads: Coding and Content.
Saai kicked things off by generating synthetic data — the backbone for our model. He mapped key metrics, identified correlations, and combined random values with known sources to mimic real scenarios.
Kshama jumped in with CSM insights, pulling merchant data like Swym revenue, session counts, wishlist contributions, and traffic metrics from Metabase to enrich the dataset.
Ram fired up the front-end with some AI-infused vibe-coding, while Adarshya added a splash of color.
Meanwhile, Kshama and Adarshya started stitching together this very blog.
📆 17 April 2025
☕ Rise and shine — 9:30 AM
Post breakfast-fuel, we dove right back in.
Ram kept building out the front-end, and Yashit hopped in with some much-needed CSS wizardry.
Saai started working on wrapping up a clean, balanced dataset and built a machine learning model using Scikit-learn, Pandas, and Numpy. Early results? Brutal — just 0.05% accuracy. But after refining inputs and tweaking the model, we pulled off a 96% F1 score.
Ram then shifted to the backend, collaborating with Raj, Mani, and Vamshi to nail down a solid design for extracting wishlist touchpoints from merchant sites — without tedious manual scraping.
Kshama and Adarshya worked on shaping our pitch deck narrative.
🎛️ Into the thick of it — 4:00 PM
Saai fetched merchant URLs from all plan tiers and got to cleaning the data from Metabase. (Believe Saai; believe us — it needed a lot of cleaning.)
Ram worked on integrating LLMs to check if wishlist essentials — like the header/launchpoint, product page buttons, and collection page options — were live on merchant sites.
Adarshya and Kshama sought comfort (and decent coffee) by the new machine and were then joined by Ram who also needed a break from things not working according to the initial plan he had.
😵💫 Existential crisis o’clock — 10:15 PM
An innocent question around “influencing factor” triggered a full-blown existential crisis.
Enter dev-god AB (read: देव = God) — who blessed us with clarity.
His gospel:
“You’re looking at revenue outcomes. Instead, look at shopper behavior. Don’t measure value backwards.”
Moral of the story: every hackathon has that moment — the soul-searching dip before the comeback.
🔄 The Pivot
While contemplating what we could realistically build in the short time we had… the real problem revealed itself.
(Okay fine — it was already one of our submitted ideas.)
The wishlist outreach idea was great. But you know what’s harder than missing buttons?
👉 Bad documentation.
Inspired by day-to-day dev struggles and unspoken frustrations, we realized docs were a silent buzzkill for developer success (or happiness?)
So, we pivoted.
🧠 New Plan: AI for Docs
Instead of flagging wishlist features, we decided to:
Scan and assess current developer documentation
Benchmark against best practices (PostHog + our own standards)
Generate smart suggestions: reorganize content, improve clarity, add visuals, shorten paragraphs, and sprinkle in code examples
Auto-generate polished documentation snippets that devs could plug right in
Docs are often the first impression, and obviously, first impressions matter.
🕛 As the clock struck midnight - 12 AM
The Cinderella moment arrived.
But unlike Cinderella, we weren’t done with the ballroom dance, we were just getting started.
The idea was thought of, but now it had to be built 🏗️
The Bellas (Adarshya and Kshama) headed back to the hotel at 12:30 AM because — like Cinderella’s carriage and magic disappearing post-midnight — their brainpower was also starting to diminish and disappear (poof 💨)
The brave heroes- Saai and Ram stayed back to code through the night.
They were joined by a handful of die-hard hackathoners who also broke into song — yes, a choir of Swymmers singing Kolaveri Di was involved. Oh and Ram doubled as a DJ for the mystical hackathon night- all attempts to fight off sleep and push toward demo day.
📅 +1= 18 April 2025
6:00 AM
The coder-bois finally called it a night (or morning?) and headed back to the hotel to rest and freshen up for D-day.
10:00 AM
After a pit stop fueling up on breakfast, we flew into work.
The documentation tool needed fine tuning and polishing.
Adarshya and Kshama hustled on redoing and completing the pitch deck to address the pivot and showcase the new tool built by Saai and Ram — and, of course, this blog.
In true Swymmer spirit, they also helped other teams with their demo prep, gave testimonials, played director to an ad shoot, and even jumped in as actors for it.
3:15 PM
After a Pizza Bakery lunch ordered in by Jahnavi, our team’s lucky charm Adarshya picked a chit that gave us our presentation number - 4th, right in the middle. Exactly what we hoped for.
All teams started giving their demos at 3.40pm
✨ Full circle
Since this hackathon was all about using AI to empower Swymmers, it only felt right to pull AI to work on this very blog too. We drafted it, but refined and tidied up this post using a trusty AI sidekick (hint: you’re reading the results).
Hackathons aren’t just about code.
They’re about detours, caffeine-fueled crises, accidental karaoke nights, and those serendipitous moments when you realize
the real problem you need to solve isn’t the one you started with.
And while we came in chasing wishlist buttons, we left with something bigger: a tool that could genuinely make life easier for the dev community we work with every day.
Would we do it again?
100%. In a heartbeat.
Because at the end of the day, it’s not just about what you build — it’s about the stories you collect, the problems you untangle, and the people you huddle with at midnight over bad coffee and great ideas.
Here’s to more pivots, Kolaveri Di karaoke, and AI-fuelled brainstorms. 🚀✨
Cheers!! 🎉
Author’s note: Hope y’all enjoyed reading it as much as we enjoyed writing it; Much love ❤️