I'm building my own prism. That's how I learned to order someone else's.
I'm Sebastián Vizzo. I run BAST as a one-person creative studio from Rosario, Argentina, working with founders and teams across the Americas and Europe.
For the past two years, I've been building Olimpo — a five-role specialized AI framework that lets me scale a creative practice without scaling a team. It runs on real briefs, in production, against deadlines.
This site is a short tour of what the system does, how it's organized, and what it has shipped.
Each role decides within its domain without asking permission for routine craft choices. The operator escalates only on irrecoverable ambiguity, real external cost, or deadlock between roles. The diagram below cycles through the run.
Directs runs, closes decisions, runs QA. No execution — direction. Translates briefs into delegations and holds the spine.
Brand voice, B2B and consumer copywriting across verticals. Writes without waiting for art — copy and art run in parallel.
Visual identity, AI image generation, motion principles. Fixes palette, type, and motion direction before development starts.
Vanilla HTML / CSS / JS. Scroll behavior, asset optimization, deploys to Cloudflare Pages. No build tools when not needed.
Senior debugging, infrastructure, integration repair. Activated when something breaks or when the architecture needs evaluation.
Cards are colored by their light from the prism — each engagement extracts its own wavelength.
Diagnostic-first sprint for eSIM.tech. Mockup + critique of the live product, framed as a partnership offer. Sent to a Senior Product Designer at the company.
esim-proposal.pages.dev ↗Interactive prototype for an AI simulation startup. Narrative inversion — the prototype is the final proof, not the opening. Sent to the founder via LinkedIn.
convergent-malabartista.pages.dev ↗A real brief used as a stress test of the agent system. The point was not the application — it was measuring how autonomously the system could deliver under a complex brief.
poly-stress.pages.dev ↗Freelance proposal for an identity design system over an existing brand strategy. Used the Brand Scout Bot to extract tokens before scoping the work.
onnodo.pages.dev ↗Same client, two pulses of work. First a tailored application landing for the role. Then, after the conversation opened, a full email campaign system developed in a single weekend — three concepts, full visual production, tone lock.
Open case study
Built on vanilla web — Cloudflare Pages, Workers, Supabase, n8n, Wrangler.
Models: Claude, Gemini, Vertex AI, nano-banana. Among others.
Scales up to React, Vite, and richer frameworks when a project warrants it.
The Shinesty engagement happened in two distinct movements. The first earned the conversation; the second showed the system working at scale. Both ran on the same brand voice lock and the same production-grade craft, but they served different purposes — and the second wouldn't have existed without the first.
Instead of a generic portfolio, a dedicated landing case study built around The Brain Hammock™ Drop. Cream-and-red palette, Recoleta serif, voice deadpan deliberate — every choice followed the internal style lock that the brand uses, not a designer's interpretation of it. The piece was the proof that the writer had already learned the brand before being hired.
shinesty.freyanode.xyz ↗After the conversation opened, a complete campaign system: Future Ancestor, Protect Your Children, and The Royal Jewels. Each one with its own palette, its own visual register, AI-generated imagery with hyper-real production, and six subject lines fixed in the brand's deadpan voice. The whole set was developed over a single weekend, end-to-end through the agent system.
shinesty-ads.pages.dev ↗Same client, two pulses. The first earned the conversation. The second showed the system working at scale.
A Strategic Sprint mockup for Bitpanda was deployed as the link inside a cold outreach email to a senior product designer at the company. Two hours after the deploy, an automated takedown notice arrived from Atelligo — the anti-phishing service Bitpanda contracts to monitor brand impersonation across the web.
The mockup wasn't phishing. It was a private design proposal, never indexed and never distributed beyond a single recipient. But the production quality was high enough that an automated system trained on the real brand flagged it as a probable scam attack.
The page was taken down voluntarily within the same hour. The proposal continued without it. But the unintended compliment stayed: the craft was strong enough that brand defense bots couldn't tell it apart from the real surface.
The lesson recorded for future runs: production-grade AI work that mirrors brand authority is detectable not just by humans — it's also detectable by automated systems trained on the same target. That is a useful signal of where the work lands.
