Our Work

We build and run AI systems ourselves.

What we teach and build for clients is tested in production — not drawn from a slide deck. Here's the proof.

Live in production

Bandorini Market App

Best Stock Market Opportunities, Twice a Week — an automated AI-analytics platform, built to detect and recover from problems fast, that scans 13,000+ global equities, narrows them to a curated ~3,200-stock universe, and surfaces the most promising BUY and SELL signals.

The problem

Analysing thousands of equities by hand, every day, isn't possible — and most investors don't have time to sift through noise to find what actually matters. The platform was built to solve this directly: automate the entire analysis pipeline, run it twice a week before market open, and surface only the signals worth acting on.

What was built

A production AI-analytics platform built solo, full-stack — from initial idea to a live product with paying subscribers. The full pipeline is automated end to end and built to detect and recover from problems fast: data ingestion, signal computation, report generation, and delivery. Built end to end using AI-assisted development tools throughout.

Python GCP Next.js Supabase Stripe AI coding tools GitHub Vercel GPT-4 API
A note on the AI layer

We're continuously improving the quantitative analysis behind the platform to sharpen the signals it produces.

Why this matters for clients

This platform is why we can offer workflow automation, data analytics, and custom AI builds with confidence — because we've designed, built, and operated a production system at scale. Bandorini doesn't just advise. We ship.

Live in production

Gift Concierge Chat Bot

A chat-style gift-finder quiz built for The Maverick Souls, a handmade jewellery brand, that matches real, in-stock pieces to what a shopper is actually looking for, then follows up by email with a personalised recommendation and a discount code to bring them back to buy.

The problem

Handmade jewellery is a considered gift, not an impulse buy. Shoppers browsing an online store often know they want to buy something meaningful, but not which piece, and a standard category page doesn't help them decide. Every visitor who leaves without a clear recommendation, or without leaving an email address, is a sale and a marketing contact lost.

What was built

A quiz embedded directly on the shop's gift-guide page. A shopper answers a few quick questions (who they're buying for, the occasion, a birth month, a jewellery type, a budget), and a rule-based matching engine picks three genuinely in-stock products with hard guarantees, such as always surfacing a real birthstone match if one exists in stock. One Claude API call per session, tuned to keep token usage and cost to a minimum while still getting the best output, writes short, personalised copy around those three products: per-product descriptions and a short "Stone Story", though it never chooses what's shown. If the shopper leaves their email, they get a personalised Gift Lookbook email with a discount code, sent via Resend. The whole thing runs as a single Cloudflare Worker, reads the shop's live product catalogue directly from WooCommerce so it's never out of sync with real stock, and is protected against bots and abuse with Cloudflare Turnstile, signed sessions, and rate limiting.

Cloudflare Workers WooCommerce Store API Claude API Resend Cloudflare Turnstile
A note on the AI layer

The AI's role here is deliberately narrow: it never decides which products to show. That's handled by a rule-based engine with hard guarantees. Claude only writes the copy around whichever three products the engine already picked, and its factual claims about each gemstone are grounded in a fixed reference dataset, not invented on the fly.

Why this matters for clients

Smaller e-commerce brands don't need a research lab, they need AI that's scoped, reliable, and actually shipped: live catalogue sync so nothing is ever recommended out of stock, bot protection and rate limiting so it can't be abused, and marketing capture built in from day one. This is what Bandorini builds for clients: production-grade, not a demo.

Live

From Brief to Live Site in a Week

This website — Bandorini's own consultancy site — built solo by Sara with no design team, no developer, and no agency. A direct proof of the AI-assisted workflow we offer clients.

The problem

Bandorini needed its own professional consultancy website — built quickly and affordably, without hiring a developer, designer, or agency, using the same kind of AI-assisted workflow we offer clients.

What was built

This website. Built solo by Sara with no design team, no developer, and no agency. The content and copy were developed collaboratively with Claude, the visual design was built with Claude, and the site itself was implemented with Claude Code — one person, AI-assisted, from first brief to a live, professional consultancy site.

HTML / CSS Claude Claude Code GitHub Vercel

Why this matters for clients

Every service Bandorini offers clients — AI training, workflow design, custom builds — was used to make this page exist. The methodology isn't theoretical. This site is the proof.

More case studies coming.

Bandorini is actively taking on new client work. As engagements complete, we'll add them here — with real numbers, honest framing, and no invented details.

Be our next case study →

Ready to build something that actually ships?

Tell us about your project — we'll come back within five business days with a free, no-obligation consultation.

Start a conversation →