In April I spent a weekend at Base Jump, an accelerator in Barcelona backed by v3v ventures, building a SaaS called BuildLoud with one teammate. The premise fits in a sentence: you built the product, let the machine market it. A solo founder types a short description of their brand, and the system plans, writes, renders, and publishes a full week of Instagram content on autopilot.
The landing page is still live at buildloud.site. This post is about what happens to your engineering when you have days instead of months, because I came out of that weekend building differently than I went in.
The problem was ours
Every product I work on has the same marketing story. Building the thing is the fun part. Then someone has to post about it, and that someone is a developer who would rather fix a bug than film a reel. One piece of content can eat an hour or two. Multiply by three posts a week and you have a part time job nobody applied for.
So the idea was not exotic. Automate the whole loop: plan the week, write the posts, render the visuals, publish on schedule, look at what worked, and adjust next week's plan accordingly. The interesting part was never the idea. It was whether two people could make the loop actually run before Sunday night.
The architecture of a weekend
We wired together things that already worked instead of building things that might. Claude sits in the middle as the planner. It reads the founder's brand brief, plans a week of content, and writes every post. Higgsfield renders the visuals with the founder's face locked across all of them, because a feed of generic AI images fools nobody, but the same recognizable person every post starts to look like a brand. The Meta Graph API publishes to Instagram on a schedule. Underneath it all is Supabase: Postgres, auth, storage, and cron jobs for the publishing calendar.
The part I ended up caring about most was the feedback loop. Every Sunday the system reads what got saves, reach, and comments, and reweights next week's plan around the hooks that actually landed. Without that, it is a content cannon. With it, it is something closer to a junior marketer that learns.
Billing was credits through Stripe, one credit for one cent, with the AI costs bundled in so users never touch an API key. Pricing AI products is its own puzzle: every generation has a real cost, and the credit math has to survive your heaviest user. We spent a surprising chunk of the weekend on a spreadsheet, not on code.
What speed does to your decisions
The rule at Base Jump was to deploy in the first hour, before writing anything of substance. It sounds like theater. It is not. Once the deploy pipeline works, every feature lands in a real environment minutes after it exists, and the question changes from "will this work in production" to "is this worth building at all."
That question got brutal fast. With months of runway you can afford maybes. With one weekend, every maybe is a no. We cut an approval dashboard, cut multi-platform support, cut everything that was not the loop itself. What survived the cutting was, almost by definition, the product.
I also learned to treat the demo as the spec. Not a slide describing the loop, but the loop running: type a brand, watch a week of content appear, watch a post go live. Working backward from that moment told us exactly which corners mattered and which ones nobody would ever see.
None of this is how I build WRDB, and that is fine. A product with real users deserves architecture, tests, and patience. But I now notice when I am polishing something nobody asked for, and that instinct came from being forced to ship with the clock running.
What survives a sprint
Honest ending: the backend is switched off now. Hackathon projects have lifespans, and keeping a full Supabase project alive for a product between lives was not the right call. The landing page stays up, the code is safe in the repo, and if BuildLoud gets a second life it will be a rebuild with better decisions, not a resurrection.
What actually survived is less visible. A working knowledge of the Instagram publishing API and its OAuth maze. A pricing spreadsheet I have reused twice since. The habit of deploying before building. And the experience of standing in front of people who fund companies, showing a thing that did not exist on Friday, and watching it post to a real Instagram account.
For a weekend, that is a good trade.