Applied AI · No hype

AI that saves real hours

We put AI to work in your business: invoice OCR, conversational agents over your knowledge, predictive analytics, automation. Direct expertise in OpenAI, Anthropic Claude and open-source models.

Real cases we already solve with AI

📄

Inbound invoice OCR

Upload the invoice, AI extracts vendor, base, VAT, total and prepares it for accounting. Saves 3-5 min per invoice.

💬

Support agent on your site

Chatbot trained with your real docs. Answers 24/7, escalates to human when stumped. Cuts easy tickets 60-80%.

📊

Predictive sales analytics

Seasonality detection, stock forecasting, churn-risk customer identification.

📧

Automatic email triage

AI classifies incoming mail, extracts key data and routes to the right department with a summary.

📝

Meeting/long PDF summaries

Drop in 200 pages, get a 2-page summary with action items identified.

🔎

Semantic knowledge search

Natural-language search over your docs. "When did we change the Endesa contract?" → answer in seconds.

Tech we use

We pick the right model for the case. We're not married to any vendor.

🤖

Anthropic Claude

For tasks needing deep reasoning and long context (1M tokens). Excellent for code and document analysis.

🧠

OpenAI GPT

For conversational generation, function calling and tooling ecosystem.

👁️

Computer vision

Pytesseract, Claude Vision, Google Cloud Vision: OCR and structured extraction.

🔬

Local models

Llama 3, Mistral, Qwen on private infra for sensitive data that mustn't leave.

📚

RAG (Retrieval Augmented Generation)

We index your docs with vectors and the model answers with cited sources.

🛠️

MCP servers

Custom MCP servers to connect AI with your databases securely.

How we build an AI project

Identify the repetitive task

We talk to you and your team. What do you do a thousand times that wastes time? That's where AI goes.

Estimate ROI

Hours saved * cost/hr - project + model cost. If it doesn't pencil, we tell you.

Prototype in 2 weeks

Minimum viable version on your real data. We try it together to see if it works.

Integrate into your flow

Hook into your CRM, ERP, email, web. Put output where it belongs.

Measure

Real metrics: % AI-resolved tasks, % errors, hours saved. No fluff.

Iterate

Tune with real usage data. If one case fails, we learn and fix it.

Real saving examples

Sector Case Metric Result
Accounting firm Invoice OCR 4hrs/week → 30 min 14 hrs/month recovered
Online shop Pre-sale chatbot 60% tickets solved no human 5 hrs/day less queue
Industry Incident triage 85% correctly classified 2 people freed up
Restaurant WhatsApp bookings AI manages calendar 24h response

How we differ

We start small

Never a 6-month AI project. 2 weeks to see if it works.

No hype

If the case isn't for AI (it's for an Excel macro), we tell you.

Your data stays home

If you want, we use on-premise models. No data leaves.

Honest metrics

We give you real hit/miss rates. No selling that everything is perfect.

Owned code

Custom implementations with official SDKs. No dependence on third-party SaaS.

Direct support

When a model changes or an API drops, we fix it the same day.

FAQ

Is AI expensive?

Depends. Invoice OCR can cost from €2,000 + a monthly model cost (€5-30/mo for normal use). Almost every project pays for itself within a year.

Is my data safe?

With Claude or OpenAI, your data isn't used for training (zero-data-retention). For maximum guarantee, we use Llama on-premise: data never leaves.

Does it work in Spanish/Catalan?

Yes, perfectly. All current models master Spanish, Catalan, English and 100+ languages.

What when the model "hallucinates"?

RAG: we force it to cite sources from your knowledge. If no source, it says "I don't know". Hallucinations nearly gone.

Can I switch models later?

Yes. Our architecture lets you swap Claude, GPT, Llama, Gemini without rewriting code. Important contractual leverage.

What if models improve?

You benefit. When a better version drops, your system uses it (or a tiny tweak). We'll tell you.

15 minutes and we'll see what makes sense for you

A short no-commitment call. If your task isn't a good fit for AI, we tell you.

Book the call