The Chilean startup utilizes massive datasets to train its AI models, which means it needs the ability to scale and access to cutting-edge tools.
Theodora uses Azure virtual machines to fine-tune its models, Azure Blob Storage to house its large datasets, Azure OpenAI for natural language processing, and Azure Functions to handle event-driven workloads.
With the help of Azure, Theodora has found more than 100,000 biases and is continuing to grow its impact.
“Even justice has bias.”
That realization led Chilean lawyer María José Martabit to found Theodora AI, a company that offers an AI-powered platform that detects and corrects unconscious bias across industries, helping companies communicate more fairly and effectively.
Martabit had been working in the legal system for 15 years and wondered how it was possible that one judge or court could come to a different conclusion than another using the same facts.
AI expert Ricardo Baeza-Yates joined as a co-founder, not just because of the technology but because of the profound impact it could have on society. Together, they are challenging the status quo, proving that AI can be a tool for justice, and building a future where decisions are driven by insight, not bias.
Because of Martabit’s background, Theodora AI initially approached bias reduction as a way to mitigate potential legal action. As Theodora AI expanded its research, it discovered that bias extends far beyond the legal system – it shapes human behavior, brand perception, and even business performance. That means communication without bias can benefit a company’s bottom line – an enticing benefit for marketing teams.
“If someone is selling jewelry for instance, they can sell more, attract more people and create a better community because of the way they communicate and because they care about bias,” Martabit says.
That discovery makes even more sense now that Theodora has done research to understand how the brain interprets bias. A Theodora Brain Lab study found that the part of the brain that is activated when bias is detected is the same part of the brain that is activated by emotions such as disgust and repulsion. The goal of the Brain Lab study was to identify and mitigate biases in language, improving decision making and organizational communication.
Theodora currently offers several tools to clients, including a bias report and a bias detector.
The report contains detailed metrics about a company’s level of bias – related to gender, age, disability, nationality, ethnicity, religion, sexual orientation, socioeconomic status or personal characteristics – as well as the impact of those biases. The bias detector identifies specific concerns and offers recommendations for remediation. Currently, the AI models that inform the bias report and detector are run internally by Theodora, and the information obtained is provided to the client. Eventually, the company plans to offer an anti-bias SaaS tool that would allow clients to do their own evaluation.
To analyze and mitigate bias at scale, Theodora AI needed advanced AI models powered by a robust cloud infrastructure. That’s why the company built its bias detection platform on Azure, leveraging AI tools like OpenAI for natural language processing and Azure Functions to handle large-scale data processing.
Theodora utilizes a massive database to train its models, which means it needed the right architecture and system to support all that data. That’s why the company has been cloud based from day one, giving Theodora the flexibility to experiment, deploy quickly and access cutting-edge AI tools without heavy upfront investments in hardware.
The bias detector platform runs on a modern web stack, with the AI models accessible via API. The system is designed to be provider-agnostic so it can support a range of customers, but Azure products and services have been crucial internally, especially given the mix of technical and non-technical experts who work at Theodora.
“I love Azure because it’s super simple and the way it works is easy to understand even for me,” says Martabit, who falls in the non-technical camp. “The control panel is easy for everyone.”
Theodora uses Azure Virtual Machines (SKU Dsv6) to fine-tune its models, Azure Blob Storage to house large datasets and model artifacts, Azure App Service for app hosting, Azure SQL for managing user data and metadata, and Azure Cognitive Services for AI tagging and content moderation. Theodora also uses Azure Functions to handle event-driven workloads. For example, the bias report needs to process heavy amounts of social media data on demand.
“The dream of Theodora was to scale everywhere, and the only possible way to do that was with technology. Technology is like a friend, so you invite this friend to help you.”
María José Martabit, Founder & CEO, Theodora AI
Theodora offers several additional tools and resources for its customers and the public. An anti-bias API tool is adaptable to any organizational requirement, and incorporates AI, bias detection and mitigation, helping to comply with regulations and internal policies. A social listening tool can monitor online conversations across different platforms, analyzing brand mentions, comments and trends, increasing awareness and mitigating reputational risks. And the company’s Cognitive Scan identifies and measures unconscious biases in people using a test adapted from the Implicit Association Test and an accessible quiz.
And a new Theodora project that will be particularly data-intensive is an art competition that asks artists to make original art that ethically integrates AI – with the goal of showing that art and AI are not antithetical but rather can complement one another. Martabit is expecting that they may receive more than a million submissions. Martabit sees the potential for AI to be applied in infinite beneficial ways.
She’s proud that her startup in Chile is already making an impact around the world. She’s particularly excited to be working with other people who care about ethics and AI, and that they’ve accomplished so much without needing venture capital. Theodora’s bias detector has found more than 100,000 biases across more than 50 organizations in 30-plus countries. By addressing unconscious bias in their communication, Theodora's clients have seen increased engagement, improved brand trust and stronger connections with diverse audiences.
Theodora is expanding its influence by opening offices in Paris this summer after receiving the Trophee Startup 2024, a prestigious award that promotes Chilean science and technology-based products with potential for international growth.
“The dream of Theodora was to scale everywhere, and the only possible way to do that was with technology,” Martabit says. “Technology is like a friend, so you invite this friend to help you.”
Follow Microsoft