FIX Wizard is Imandra’s first AI assistant to be deployed across its product line. Built on Imandra’s novel neuro-symbolic architecture, FIX Wizard automatically reasons about complex financial protocols and APIs, acting as an expert AI assistant for onboarding clients in capital markets.

FIX Wizard combines statistical AI, used by LLMs, and Imandra’s automated reasoning to create a powerful Generative AI assistant grounded in logic, with independently verifiable audit trails that meet stringent regulatory requirements. This approach addresses the fundamental shortcomings of applying LLMs in regulated markets, which include hallucinations, lack of scalability to unseen inputs and lack of validation of knowledge sources.

"LLMs hold tremendous promise, but ultimately cannot be trusted in regulated environments. By combining their strengths with scalable, rigorous automated reasoning, we obtain a kind of magic: conversational interfaces with correct reasoning and domain-specific skills," said Dr Grant Passmore, Co-CEO of Imandra.

Today, capital markets run on a myriad of complex and interconnected trading systems. Historically, navigating PDF specifications of FIX interfaces was highly manual and error-prone. The industry has been working on initiatives like FIX Orchestra - a digitized representation of FIX specifications. Imandra takes this approach further by creating a full logical model of the API.

FIX Wizard uses Imandra’s "digital twin" of a FIX gateway and automated reasoning to:

• Answer questions about the FIX specification and rules of engagement

• Analyze customer-provided FIX traffic, understand the underlying causes of issues and make recommendations on how to remedy them

• Diagnose many issues at once, not just feeding back the first problem encountered

• Give guidance on certification test cases and run conformance tests against the implementation

"LLMs are good at translation and bad at reasoning. Automated Reasoning is bad at translation (i.e., requires you to be very precise) but incredible at reasoning. When we combine these techniques we get ground-breaking products that can be safely deployed in regulated environments," said Denis Ignatovich, Co-CEO of Imandra.

How it works:

Imandra’s interface with LLMs builds on years of Imandra’s use within financial markets and government and defense. Imandra’s underlying technology is capable of logical reasoning about "real-world" algorithms including complex financial trading systems, collision avoidance algorithms and neural networks. This scalability is the result of numerous scientific advances leveraged by the product and its cloud-native design.

Imandra’s neuro-symbolic architecture alleviates LLM woes like hallucinations and incorrect "statistical reasoning." Imandra allows users to encode complex formal "mental models" of entities like API specifications, spreadsheets, knowledge bases, financial models, and more. The LLM then uses Imandra as a "reasoning assistant" where it formalizes its "thoughts" and asks Imandra to analyze their validity (e.g., with respect to regulations) and to perform complex reasoning for answering questions and synthesizing results.

Today, Imandra is working on several "Wizards" combining its reasoning engine with LLMs - including for SysML - a language used in Model-Based Systems Engineering (MBSE) for designing complex systems in avionics, defense, automotive and other safety-critical industries.

The company has also released a Python API to allow others to easily build AI assistants incorporating Imandra’s automated reasoning.