We build software with AI at its core.
Products, platforms and specialised machine learning, designed, engineered and run in production by a senior team that has been shipping software for 15+ years.
Prototypes are easy. We build the version that survives real users, real data and an audit.
building software for startups and enterprises. Today, AI-native products end to end.
Most AI products stall after the demo.
The model is rarely the problem. What breaks is everything around it.
Getting a model to answer well once is easy. Getting it right ten thousand times, on real data, at a predictable cost and with a record of every decision is an engineering problem. These are the failure modes we design out from day one:
- No evaluationWithout an evaluation suite, a new model version or prompt quietly makes results worse and nobody notices.
- No traceabilityIf an answer can’t be traced back to its source document or data, it can’t be audited, explained or fixed.
- Unplanned cost and latencyInference that costs cents in a demo can cost thousands a month in production. We model it before we build.
What we build.
From the first screen to production, with machine learning where it actually moves the needle.
- 01
AI-native products
Web, desktop and mobile products where AI is the core feature, not an add-on: product design, full-stack engineering and cloud infrastructure in one team.
- 02
AI agents & RAG systems
Multi-agent workflows, retrieval-augmented generation and MCP integrations with your tools, built with guardrails, evaluation and human-in-the-loop.
- 03
Applied machine learning
Custom models when off-the-shelf isn’t enough: speech-to-text, document AI, scoring, ranking, forecasting and decision models trained on your data.
- 04
AI inside your existing product
New ML and LLM features shipped into live products without stopping what already works, from a single model to a full agent.
- 05
Web & desktop software
Not everything needs AI. Custom web platforms, SaaS and internal tools on AWS, Google Cloud or Azure, and desktop apps for macOS, Windows and Linux, engineered with automated testing and continuous delivery, as we have since 2009.
The engineering underneath.
Every technical term, explained in one sentence, so you know exactly what you’re getting and why it matters.
Cloud & MLOps
We deploy where your company already runs, with infrastructure as code, containers and CI/CD, so every environment and every model release can be reproduced and audited.
AWS (Bedrock, SageMaker, Lambda) · Google Cloud (Vertex AI) · Azure · Terraform · Docker
Agent harness
Everything around the model that keeps an agent reliable in long tasks: context management, verifiable steps, retries, persistent state and evaluation of every output.
LangGraph · multi-agent orchestration · tool calling · MCP · guardrails
Decision models & classic ML
Systems that score, classify, rank or recommend by combining business rules, classic machine learning and LLMs. Every decision can be explained and reviewed.
Scoring · classification · ranking · deterministic logic · explainability
RAG & semantic search
Models that answer from your own documents rather than from memory, and cite the source of every answer so it can be verified.
Embeddings · vector databases · Document AI · hybrid search · reranking
Custom & open models
When a general model isn’t enough, we adapt it to your domain or run open models on your own infrastructure or on-device, so sensitive data stays in-house.
Fine-tuning · SFT · DPO · LoRA · open-weight models · local inference
Evaluation & observability
Every step of every run is traced and scored, so we know what fails, what it costs and how long it takes before your users do.
Tracing · Langfuse · automated evals · trace scoring · cost and latency
New product or already in production?
New AI product
From idea to MVP to scale: a web, desktop or mobile product with AI at its core, validated on real data before you commit to the full build.
How we workAI inside your product
We add machine learning and LLM features to products already in production, without stopping what works. A model that helps people decide, or an agent that does the heavy lifting.
Validate.Ship.Scale.
Every phase ends with a decision: continue, change course or stop.
- 01 · Proof of concept
We test the idea on your real data. Deliverable: a technical report with measured accuracy, cost per use, latency and risks.
- 02 · MVP
A first version in production, with code in your repository, CI/CD, monitoring and documentation from the first sprint.
- 03 · Scale
Load testing, observability, model spend control and an evolution plan for when AI providers and models change.
Built to pass an audit.
Compliance isn’t bolted on at the end. It’s part of the architecture from the first line.
EU AI Act
We classify your system’s risk level and prepare what the AI Act requires: technical documentation, logging, transparency and human oversight.
GDPR and data
Data minimisation, impact assessments, EU data residency and models that are never trained on your information.
LLM security
Testing against prompt injection and data leakage following the OWASP Top 10 for LLMs, access control and hard limits on what an agent can do.
Traceability
Every decision the system makes is logged and auditable. If a regulator asks, you have the answer.
- EU AI Act
- GDPR
- OWASP LLM Top 10
- NIS2
Selected work.
All case studiesOne platform for operations, ticketing and loyalty across multiple venues.
Domain-tuned speech-to-text and supervised scoring models for compliance and sales, in real time.
Recent work also includes a multi-agent system for clinical triage, a clinical RAG app and an AI desktop copilot for meetings.
A senior team behind every project.
Software engineering, applied ML, infrastructure and product, with decades of combined experience shipping to production.
Common questions.
What does an AI product development company do?
We design and build software where AI is the core: from the product and its interface to the models, data pipelines and cloud infrastructure behind it, and we run it in production with evaluation and monitoring.
Can you add machine learning to an existing product?
Yes, it’s a large part of our work: ML models and AI agents integrated into products already in production, such as real-time call scoring or a risk-analysis agent for a fintech, without stopping what works.
Do you build custom models or only use LLM APIs?
Both, depending on what the problem needs. We use commercial LLMs on AWS, Google Cloud or Azure, fine-tune or run open models when data must stay in-house, and train classic ML models for scoring, ranking or forecasting.
Do you comply with the EU AI Act and GDPR?
We design every system with the AI Act and GDPR in mind from the start: risk classification, documentation, traceability, human oversight and data hosted in the EU.
Where are you based?
In Barcelona. We work in English, Spanish and Catalan with clients across Europe and the US.