Tekkare's Solution: Four Pillars for Reliable AI

This article is part 3 of a 4-part series on the barriers and solutions to building reliable AI in healthcare.

The first two parts of this series highlighted the main barriers slowing down AI adoption in healthcare: data fragmentation and lack of traceability, the limits of language models for critical tasks, and infrastructure constraints. These challenges show that raw AI remains insufficient to meet the requirements of such a regulated and complex environment. From this third article onward, we shift to a solution-oriented perspective. We introduce the key pillars of Tekkare’s approach, designed to build AI agents that are truly reliable, usable, and aligned with business needs.

 

OIP Databank®: The Foundation for Specialized AI Agents in Healthcare

At the core of Tekkare’s approach lies OIP Databank®, one of the largest open and structured health databases in the industry. It centralizes over 500 public data sources and transforms them into more than 1,000 cleaned datasets, normalized, enriched with business metadata (therapeutic area, status, indication, etc.), and made interoperable with existing analytical tools.

 

Designed for the real needs of data teams, developers, and AI, OIP Databank® provides reliable, traceable, and contextualized data, essential conditions for enabling AI agents to operate in a strict analytical and regulatory framework.

 

The Databank is the foundation of OIP Healthcare & Discovery: an ecosystem where data is structured to automate, document, and secure the use of AI agents serving pharma stakeholders. From this foundation, the four pillars of our approach unfold to build AI that is truly useful, reliable, and compliant.

 

Pillar 1: Unified, Structured, and Actionable Data

 

The challenge: turning the mass of open data into usable resources

Having the right data at the right time has become a critical issue in healthcare. The richness of public databases collides with operational reality: data is dispersed, heterogeneous in format, poorly aligned semantically, and difficult to query transversally.

In an environment where precision and traceability are critical, the quality of input data directly conditions the reliability of downstream analyses.

 

Our approach: a data infrastructure designed for healthcare use cases

Tekkare built OIP Databank® specifically to address this barrier. Since 2015, we have developed a dedicated infrastructure for consolidating and structuring public health data. This platform unifies heterogeneous sources into a single, connected, traceable, and AI-ready base.

It includes:

• Regulatory databases (EMA, FDA, HAS, BDPM),
• Therapeutic references (ATC, ICD-10, Orphanet),
• Scientific publications (PubMed),
• Economic data (public prices, reimbursement rates, etc.).

This structuring does not rely solely on algorithms. It is reinforced by manual curation processes: removing redundancies, aligning nomenclatures, enriching with business annotations, and integrating critical metadata. This constant dialogue between automated tools and human expertise ensures a database that is not just technically “clean,” but genuinely usable for pharma use cases.

 

Concrete example: Price Tracker

One of the most telling use cases is multi-country price monitoring. Through the Price Tracker application integrated into OIP Healthcare®, pharma teams can:

• Track the evolution of public prices in over 20 countries,
• Filter data by molecule, manufacturer, brand, country, or generic status,
• Compare price levels with full context (price increases, upcoming prices, new products),

This type of tool allows teams to shift from manual monitoring to structured, actionable intelligence, directly reusable in deliverables. This pillar of structured data is therefore the first building block for building truly useful AI agents in a regulated, complex environment.

 

Pillar 2: Semantic Models to Guide Business & Scientific Analysis

 

The challenge: correctly interpreting complex and variable concepts

Even with a well-structured database, business understanding remains a major obstacle. In healthcare, the same term can refer to very different realities depending on the country, regulatory authority, or context of use (price, indication, reimbursement, formulation, etc.). For example:

• An indication may be phrased differently by the EMA, HAS, or FDA,
• A rare disease may have an Orpha code, an ICD-10 code, or be described in free text,
• A price may refer to very different regimes (FSS rate, Big4, public retail price, etc.).

Without interpretation models, even the best AIs struggle to connect the right dots, prioritize the right information, and deliver context-appropriate answers.

Semantic models do more than standardize vocabulary. They act as semantic arbiters, correcting inconsistencies, highlighting regulatory nuances, and preserving business & scientific logic in highly regulated environments. They also serve as filters against implicit biases in raw data by encoding explicit rules on what can (or cannot) be compared.

 

Our approach: semantic models designed for the pharmaceutical industry

We have developed semantic models that act like intelligent business maps of the pharmaceutical domain.

They establish explicit links between key industry concepts: drug, indication, regulatory status, price, target population, etc. Concretely, these models enable AI to:

• Automatically link a publication to a therapeutic area, ATC class, and marketing authorization status,
• Prioritize information based on the level of evidence or regulatory source,
• Contextualize answers according to usage (pricing, monitoring, benchmarking, etc.),
• And most importantly, avoid fundamental errors caused by language misinterpretations.

 

Concrete example: Global Vaccination

The Global Vaccination application, integrated into OIP Healthcare®, allows users to compare vaccine access conditions across countries, by:

• Age group,
• Gender,
• Number of doses administered (coverage rate),
• National vaccination strategy (vaccination schedule).

Thanks to the semantic models embedded in OIP, the AI agent can:

• Recognize different versions of the same vaccine across countries (e.g., commercial names),
• Rapidly compare vaccination strategies country by country,
• Structure the data in a homogeneous way,
• And, most importantly, document the regulatory sources used: issuing body, publication date, authorization status, etc.

This interpretation layer is essential to ensure that AI-generated analyses respect industry logic, avoid approximation biases, and deliver outputs that are immediately usable without manual reprocessing.