Oxide AI is a startup that researches and develops advanced large-scale AI technology frameworks combining computational and generative AI (gen AI) models. Its focus is the financial sector, and its first product launch is Oxogen, a B2C product aiming to redefine the investment scene with a disruptive service for modern investors.

The firm’s vision is to go beyond the limitations of traditional stock screeners and offer investors an efficient way to stay informed and discover new financial opportunities. It robustly resolves the complexities of analyzing thousands of companies, going beyond simple market data samples—a task that far exceeds human capabilities. Oxogen’s AI models use both quantitative data (such as price and volume) and qualitative evidence (such as patents, research papers, news feeds and social media) to bring high-value insights without information overload. What makes Oxogen unique is its ability to fully integrate reasoning based on numerical information and gen AI, delivering explanations and transparency to support the information it surfaces.

In upcoming releases, Oxogen will enable users to design their own automated research missions, providing personalized AI-supported investment research. Oxide AI also plans to bring technology aimed at the B2B enterprise applications space with a more extensive support for automated deep AI-research of whole financial markets.

In the proof-of-value (POV) development of its product, Oxide AI worked with IBM^® Client Engineering to pilot IBM watsonx.ai^™, IBM’s studio for gen AI and machine learning. Oxide AI’s vision was to replace its use of proprietary large language models (such as OpenAI and ChatGPT) and disparate cloud infrastructure. Its objectives were to improve operational stability, increase control over its gen AI models and actively participate in a growing ecosystem of transparent and more task-specialized models. The watsonx.ai platform is designed to bring all the necessary building blocks into one coherent service, enabling teams to go from an idea to a robust deployed system with minimal friction.

The pilot demonstrated very promising outcomes:

• Within less than a month of utilizing watsonx.ai, Oxide AI met a 95% qualitative acceptability threshold for the content it generated.
• The solution achieved 37% faster average sequential response times than OpenAI GPT-4, while maintaining comparable quality levels.
• Oxide AI decreased its carbon footprint with simpler AI model implementation that reduced the need for fine-tuning.

“In a dynamic co-creation phase, our team has successfully integrated watsonx.ai’s gen AI into our existing Oxogen solution currently running on OpenAI’s ChatGPT. In IBM watsonx.ai we are utilizing the open-source model Llama 2 and this is a significant step for us, since we were looking for an open alternative that could deliver on a similar quality and with potential to be used in enterprise-grade production environment. We are now working towards fully incorporating watsonx.ai’s gen AI into Oxogen’s production pipeline. We are confident that this tight integration, coupled with custom adaptations in line with our data processing requirements, will yield substantial performance boost. The next step of collaboration between IBM and Oxide AI involves specialized language models taking advantage of Oxide’s unique multi-year financial data assets that will drive more value and competitive advantage than generalized LLMs.”
—Lars Hard, CEO/CAIO, Oxide AI

“IBM watsonx.ai has an intuitive, secure workflow that enables immediate access to major open-source models, which were primary candidates for us to test and evaluate. The platform has been effectively utilized by both data scientists for more extensive testing and experimentation via Python SDK, as well as by the content team, who were able to swiftly complete prompt engineering and configurations to assess output quality using watsonx.ai.”
—Kateryna Wikstrom, Head of Product, Oxide AI

“IBM Client Engineering team has been very instrumental in navigating us through the vast landscape of open-source models, providing us guidance and helping us pinpoint the most efficient approaches. They have also facilitated models benchmarking to ensure we select the most suitable model, in both output quality and scalability potential.”
—Kateryna Wikstrom, Head of Product, Oxide AI