MIG PRISMA and MIG EMBRYONIC create ORIENTIA for SANTALUCÍA.

We have developed an artificial intelligence platform for the Santalucía Institute—specializing in retirement, pensions, and social welfare—that transforms nearly a decade of expert knowledge into accessible, contextualized answers backed by highly authoritative sources.

Artificial intelligence is transforming far more than just the technology organizations use. It is changing how people search for information, phrase their questions, and expect to find answers. Moving beyond a digital logic built over years around search engines, web pages, and documents, we are advancing toward a scenario where the user asks questions directly and expects a response that is understandable, contextualized, and reliable enough to inform further decision-making.

That was the starting point of ORIENTIA, the new conversational artificial intelligence platform that we have created at MIG PRISMA and MIG EMBRYONIC for the Santalucía Institute, an entity belonging to the Santalucía Group dedicated to research, dissemination, and knowledge generation in areas such as retirement, pensions, savings, longevity, and financial education.

The challenge lay in transforming years of expert knowledge into a new way of accessing that information, without compromising the rigor, authority, and trust upon which the Santalucía Institute has built its standing.

A conversational experience

Over the years, the Santalucía Institute has built up a wealth of knowledge comprising studies, reports, barometers, publications, and contributions from its Expert Forum. This is a body of information of immense value which—as is often the case with knowledge-intensive organizations—was originally designed primarily to be read, consulted, or located via traditional navigation and search models.

The work carried out during these months has consisted of prepare that knowledge so that it can be interpreted, retrieved, and utilized by a generative artificial intelligence architecture.

This has entailed analyzing and structuring the Institute's document corpus, establishing criteria for information retrieval and prioritization, defining the subject areas ORIENTIA can address, and building an architecture capable of locating relevant knowledge before formulating each response.

But self-knowledge doesn't operate in isolation. ORIENTIA can compare and complement it with a pre-authorized ecosystem of institutional sources and highly reputable entities, incorporating regulatory, statistical, and documentary information from official sources. In this way, the system combines knowledge generated by the Institute itself with authoritative external information, thereby enhancing the currency, context, and rigor of its responses.

An AI that does not respond based on the model's open knowledge.

This is probably one of the fundamental differences of the project. ORIENTIA uses an architecture of Retrieval-Augmented Generation (RAG) combined with mechanisms for retrieving up-to-date information from previously authorized external sources. When a user asks a question, the system does not simply pass it on to a generative model to obtain an answer.

Before responding, the architecture identifies the query's intent, retrieves relevant knowledge fragments, selects and prioritizes the most pertinent information, and constructs the documentary context upon which the generative model subsequently operates.

The result is an AI whose behavior is conditioned by a defined and governed knowledge perimeter, especially important when dealing with matters such as retirement, pensions, social security, or financial planning, where a seemingly plausible but incorrect answer can be far more problematic than the absence of an answer.

Control layers

A significant aspect of the project has focused precisely on defining its boundaries. ORIENTIA has been designed as a specialized information and outreach tool. It does not provide financial, legal, or tax advice; it does not prescribe products; and it is not intended to replace the judgment of professionals or competent public authorities.

To maintain this positioning, we have worked on various control layers governing the query, knowledge retrieval, and response generation. The system incorporates intent classification, thematic scoping, controlled source selection, system instructions, and mechanisms designed to reduce unsubstantiated responses, handle out-of-scope queries, and thwart attempts to manipulate the assistant's instructions.

Because developing AI for an organization is not just about getting a model to respond. It consists of determining under what conditions it can respond, using what knowledge, with what limits, and with what guarantees.

Technology, but also language

There was another, less visible—yet essential—challenge. A technically correct answer can still be a bad answer if the user does not understand it.

That's why we've developed ORIENTIA from a perspective where technological architecture and conversational experience are part of the same problem. The system incorporates specific criteria for plain language (plain language) to translate technical, regulatory, or financial concepts into accessible explanations without sacrificing the necessary rigor.

Technology makes it possible to retrieve knowledge. Conversational design must ensure that this knowledge proves useful.

From strategy to actual deployment

ORIENTIA also represents the evolution we are driving in MIG's technological capabilities.

The project has required integrating disciplines that could traditionally be addressed separately: strategy, content, technology, data, UX, artificial intelligence, security, privacy, and governance.

Through MIG PRISMA and MIG EMBRYONIC, we have worked on the solution's functional design, technology architecture, knowledge preparation and ingestion, retrieval and re-ranking systems, component orchestration, generative model integration, control mechanisms, conversational memory, and the infrastructure required to deploy the system into an operational environment.

The developed architecture uses technologies of RAG and Live RAG, vector databases, and orchestration systems, combining different components and technology providers to prevent the solution's value from depending exclusively on a single foundational model.

However, the most significant capability resulting from the project is likely not any specific technology. It is the development of a methodology for transforming complex corporate knowledge into specialized artificial intelligence systems that are controlled and usable in real-world environments.

ORIENTIA is a unique case. The opportunity is far greater.

Our expert AI know-how is applicable to many other organizations facing a similar challenge: they possess vast amounts of knowledge—documentation, regulations, procedures, research, products, manuals, content, or knowledge bases—yet accessing it still depends on knowing where to look and how to interpret it.

The same technological logic can be applied to internal knowledge assistants, employee support tools, customer service platforms, specialized assistants for sales or mediation teams, document retrieval systems, or experiences capable of guiding the user through complex processes.

It can also evolve. An architecture initially conceived to provide information can, in other contexts, connect with systems, services, data, or product catalogs and—subject to the appropriate controls and requirements—become a tool capable of supporting processes much closer to the business.