Projects & Case Studies

Real business problems, solved with AI, web and software.

No logos here. You'll see something more useful: the kind of challenge we solve, what we build, and the impact you can expect, backed by industry data.

A typical system we deliver
Input
Your business channels
Process
AI + your own rules
Output
Action in your systems

What we know how to build

AI customer-service agentsLLM-powered automationDocument data extractionCRM & ERP integrationsLead-generation sites in Next.jsCustom B2B portalsUX/UI designThree.js & WebGLTechnical SEO & analytics
Case studies

Four challenges, four systems.

Representative scenarios of the challenges we work on. The figures are industry benchmarks, with sources, not the results of a specific client.

Booking assistantWhatsApp · 24/7
Hi, do you have a physio slot on Thursday afternoon?
Yes. On Thursday I have 5:30 pm or 6:15 pm with Marta. Which works better?
5:30 pm, perfect.
Booked: Thursday 5:30 pm. I'll send you a reminder the day before.
checkAppointment created in the clinic's calendar
Physiotherapy clinic · MadridApplied AISample scenario

A front desk that never leaves an appointment unanswered.

Two receptionists, a phone that never stops and dozens of WhatsApp messages left unanswered after hours. Every lost message is a patient going to another clinic.

The challenge

Answer and book without hiring, and cut the appointments nobody shows up to.

What we build

An AI assistant on WhatsApp that understands real availability, books directly into the clinic's calendar, sends reminders and hands over to a person when the conversation needs one.

Industry benchmark impact

−25–40%

no-shows with automated reminders (typical range)

38% → 24%

no-show rate before and after in a clinic study (US)

Sources: Am. J. Med. (reminder-systems review); pediatric clinic study via RingRX

Scope: Flow design, AI agent, practice-software integration, reminders

WhatsApp Business APILLM with business rulesCalendar sync
Input · Email
invoice_0472.pdf
AI extraction
SupplierSuministros Norte SL
Net€1,240.00
VAT 21%€260.40
Account600 · Purchases
Output · Accounting
0472Posted
0471Posted
0470Review
0469Posted
Tax & accounting firm · MadridApplied AISample scenario

Invoices that book themselves, with a person reviewing what matters.

Hundreds of invoices a month arriving by email in different formats. The team spends the first hours of every day reading, typing and classifying them.

The challenge

Remove manual entry without losing human control over doubtful cases.

What we build

A workflow that reads every incoming invoice, extracts the data with AI, proposes the ledger account and creates the entry in the firm's accounting software. Anything below a confidence threshold goes to a review queue.

Industry benchmark impact

70–80%

of invoices with no manual intervention after 12 months (typical range)

−70%

cost per invoice versus manual processing ($10.18 → $2.94)

Sources: Serina (STP rates); APQC benchmarks cited by Ramp and Artsyl

Scope: Process audit, AI extraction, accounting integration, review queue

Document extractionConfidence thresholdERP integration
Request a quote
Qualified lead
Industrial installations · B2BWebSample scenario

From a website that just "exists" to one that asks for the quote for you.

An installations company with twenty years of track record and a website that doesn't generate a single enquiry a month. The architecture is rebuilt around one action: requesting a quote.

What we build

A Next.js site with a page per service type, a qualification form connected to the CRM, and technical SEO from day one.

Industry benchmark impact

2–5%

of visitors requesting a quote: a realistic B2B target (median 2.9%)

−14%

bounce rate for every second of faster loading (LCP)

Sources: Surface Labs and Martal (B2B conversion); MigrateLab (speed and bounce)

Scope: Strategy, UX/UI, Next.js development, CRM integration, SEO

Next.jsQualification formTechnical SEO
CustomerOrderStatus
Ferretería Ochoa#2418Confirmed
Almacenes Duero#2417Confirmed
Suministros Vega#2416Preparing
Orders through the portal · last 7 weeks
Wholesale distributor · Madrid regionSoftwareSample scenario

An ordering portal that replaces WhatsApp and phone calls.

Recurring orders from more than a hundred customers arrive by phone, WhatsApp and email, then get retyped into the ERP. Errors every week, and a sales rep tied to the phone.

What we build

A B2B portal with per-customer catalogue and pricing, repeat orders in two clicks and direct sync with the ERP. Nothing to install: it works from a phone.

Industry benchmark impact

−40–60%

phone and email enquiries after launching the portal (typical range)

−60–80%

order errors compared with manual entry

Sources: Aberdeen Group via HumCommerce; WizCommerce

Scope: Product design, full-stack development, ERP integration, training

B2B portalERP integrationPer-customer pricing
Lab

What we build when nobody asks us to.

Our own explorations, with no client. This is where we test 3D, motion and interface techniques before bringing them to a real project.

Ascent Studio — interactive 3D mountain experience
Interactive 3D experience

Ascent Studio

A mountain that changes season as the user explores it. Three.js and WebGL.

Experience itopen_in_new
Demon Drop — cinematic WebGL experience
WebGL experience

Demon Drop

Cinematic camera, particles and real-time lighting for a fictional product.

Experience itopen_in_new

Facing a similar challenge?

Tell us about it. We reply within 24 hours with an honest assessment of whether we can help, and how.