Giskard's collaborative open-source platform helps developers build better AI products

Giskard is a European startup working on collaborative, open-source software that ensures the quality of AI models by automating and simplifying several aspects of the testing process. Motivated by the obstacles that riddle the ML testing process, co-founders Alex Combessie, Jean-Marie John-Mathews, and Andrey Avtomonov decided to combine their experience and apply it to the development of a platform that would enable developers to overcome the obstacles faced in the ML testing process and ensure that their products are regulation-compliant.

This is especially important, seeing that the EU and several other jurisdictions are set to enforce AI regulations, with non-compliance leading to hefty fines in some cases. According to the startup's website, non-compliance to new AI regulations can cost up to 6% of a company's revenue. The main issue uncovered by the team at Giskard is that working toward compliance is usually a time-consuming, hard-to-scale affair, with tests and reports frequently created manually, leading to inconsistent adoption across projects and teams. Even when these obstacles are averted, there is no guarantee that a testing process will cover the full range of risks faced by AI models and their applications.

Earlier this year, Giskard received a 3 million EUR strategic investment from the European Commission to develop a SaaS platform to automate compliance with the upcoming EU AI Act. Giskard's framework stemmed from that goal and currently consists of three elements: the ML Testing Library, the AI Quality Hub, and the LLM monitoring tool LLMon. The ML Testing library is a Python library that can be integrated into any LLM retrieval-augmented generation project. An essential feature of this library is that it is compatible with popular tools such as Hugging Face and PyTorch, TensorFlow.

The AI Quality Hub is Giskard's paid offering and is an application built with scalability in mind. The Hub features advanced ML testing capabilities like visual ML debugging, a centralized dashboard, interactive insights, model comparison, and human feedback collection. It also features a ready-made customizable test catalog with over 100 functions. The Hub can be implemented on-premises or in cloud hosting services. Giskard plans to integrate its AI compliance features into this product.

Finally, Giskard's latest product is an LLM monitoring tool called LLMon, currently available as an early access beta. LLMon is a real-time monitoring tool that helps developers protect their LLM applications against common risks such as hallucinations, toxicity, and lack of robustness. Although LLMon currently only integrates with the OpenAI API, the team is actively working on integrations with the most popular LLM hosts and providers, such as Hugging Face, Google, Anthropic, Cohere, and Meta.

Overall, it seems that Giskard is in a favorable position to become a go-to solution for AI testing and compliance needs. The team's combined experience and knowledge of regulation, especially of the EU AI Act, together with their mission to create an open-source collaborative platform that is customizable and compatible with all the major LLM providers' APIs, should convince anyone that Giskard will become one of the top players in the field.