Production-Ready Custom AI Development
We design and deliver production-ready AI solutions (RAG, assistants, and automation): auditable, grounded in your own knowledge sources, and with your data securely hosted in Europe.
Most AI Projects Never Make It Past the Pilot Stage
A demo impresses in a meeting, but never makes it into production. The model confidently provides answers that are simply wrong. Company data ends up in third-party services outside your control. And when a client or auditor asks why the system made a particular decision, there is no answer.
The gap between a demo and a system you can trust lies in the engineering around it. That’s where we work.
AI with Governance: Traceable, Auditable, and Hosted in Europe
Built for the European Regulatory Framework
Every system we deliver includes governance documentation—covering data sources, human oversight, and traceability—ready to support your vendor approval and compliance processes, including considerations related to the EU AI Act. This helps ensure your AI system does not become a regulatory challenge further down the line.
What We Build
With those principles in mind, this is what we build.
Enterprise RAG and Assistants Built on Your Knowledge
AI assistants that provide answers based on your internal documentation, contracts, manuals, and catalogs, with every response grounded in verified source data.
AI Agents and Process Automation
Chatbots and Conversational Assistants.
For customers and internal teams, seamlessly integrated with your existing business systems.
AI Grounded in Your Data
Classification, data extraction, semantic search, and recommendation systems built around your organization's own information.
AI Integration for Existing Applications.
We integrate AI into existing production systems without requiring a complete rewrite.
How We Work
We start with the use case and validate its technical feasibility before making any promises.
We build a proof of concept with clearly defined, measurable success criteria agreed upon from the outset.
We develop iteratively, validating quality and performance at every step.
We deploy it to production with the appropriate metrics and oversight for your specific use case.
Technology
We work with the model that best fits each use case rather than locking ourselves into a single provider. Whether it's OpenAI, Anthropic, Google, or open-source models, we choose the most appropriate option for the job. We can also evaluate the use of local models when data sovereignty, compliance, or performance requirements call for it.
On top of the model, we build a RAG architecture using PostgreSQL and pgvector, allowing us to combine vector search and relational data within a single database. We also implement an evaluation layer with fidelity and relevance metrics based on a curated test dataset, monitored through self-hosted tools running on European infrastructure.
Finally, we integrate everything into your applications using the same engineering practices we apply to the rest of your software stack: Symfony, Google Cloud, and Infrastructure as Code. We build AI on top of production systems with the same standards of reliability, maintainability, and engineering discipline as any other business-critical application.
AI Applied in Completely Different Contexts
Four production systems, each using AI in a different way: classification, recommendation, image generation, and conversational search.
Nomadia
Nomadia is a platform for digital nomads. The entire process of discovering destinations, events, and activities is powered by a chatbot that matches vectorized catalog data with each nomad’s interests and the destinations they have already visited, in order to recommend new places to explore.
Saisho
Customers send a photo of the room where they would like to hang an artwork, and the advisors use AI to generate a realistic rendering of how each piece would look in that actual space. This allows customers to visualize the artwork in their own living room before making a decision.
Patrimonio cultural
Spain’s largest archive of historical posters. Premium users access the collection through a chatbot that understands natural language queries and translates them into the catalog’s internal filtering structure. Using those same posters, the AI can also generate educational resource packs that help teachers prepare their lessons.
Why Softspring
Real Engineering Behind AI
We build complete systems, with everything that entails in a production environment.
We Take It to Production
What we showcase works in the real world, with real users.
European Standards
Governance, traceability, and EU-hosted data as standard.
15 Years Building Reliable Software
Our AI solutions are built on that foundation of software engineering expertise.
Frequently Asked Questions
How Much Does an AI Project Cost?
We bill by dedicated person-weeks, which means costs are fully transparent: you always know exactly what you're paying for. We typically start with a pilot involving a minimum of two weeks of dedicated work—enough to deliver something tangible and validate the approach before deciding on the next steps.
From there, we scale the project according to your specific needs and objectives.
Will the AI Make Things Up?
With a well-designed RAG system, responses are grounded in your own sources, and the system is able to recognize when it does not have the information instead of improvising. We also implement an evaluation layer that measures whether responses remain faithful to your data.
This dramatically reduces hallucinations, although it cannot eliminate them entirely. That is why, for sensitive decisions, we always keep a human in the loop.
Do My Data Leave the Company?
We design our solutions so that they don't. We work on European infrastructure and, where required, ensure that all processing remains within the EU. We sign a Data Processing Agreement (DPA) and implement the controls appropriate to each use case.
In some projects, sensitive information never even reaches the model. For example, in La Psicóloga en Casa, only the described issue is processed, while the patient's personal data remains outside the AI system.
Does This Comply with the EU AI Act?
We build with this in mind from the design stage: documentation of data sources, human oversight, and traceability, ready for your clients' vendor approval processes. Most projects do not fall into the "high-risk" category, so the primary obligation is transparency.
We do not issue legal certifications—that is the responsibility of a legal advisor or, where applicable, a notified body—but we deliver a system that is prepared to undergo those reviews successfully.
How Much Data Do I Need to Get Started?
Fewer than you might think. RAG makes use of the documentation you already have—manuals, datasheets, historical records, and catalogs—without requiring you to train a model from scratch or build a large labeled dataset. The quality and structure of the information matter far more than the quantity.
How Soon Will I See Results?
We structure our work around an initial proof of concept, so you have something working in front of you within weeks, not months. In most cases, you can see real results on your own data within 2 to 4 weeks.
Do you have a use case in mind? Tell us about it, and we'll tell you whether AI is a good fit and how to make it work.
Tell Us About Your Use CaseThis is how we think, this is how we tell you.
-
AI Governance in Companies: How to Comply with the EU AI Act Without “Black Boxes”
AI is no longer experimental: companies want to integrate it without taking on operational, legal, or reputational risks. With new European regulations, “test and see what happens” is no longer enough. The key is AI governance — building systems with traceability, human oversight, and control from the start to ensure security, compliance, and sustainable growth.
-
How much does AI cost
I have asked DALL-E to generate a self-portrait of an AI, for this post, and the result is the image that heads the article... a bit weird, to be honest, a mix of thinker and mannequin, but that's what it has generated without much “prompts engineering”, as they call it nowadays. It's the only thing I've asked an AI to do for this article, on the other hand.