Denis Ignatovich, co-founder and co-CEO of Imandra, has more than a decade of experience in trading, risk management, quantitative modeling and design of complex trading systems. Prior to founding Imandra, he led Deutsche Bank’s Central Risk Trading Station in London, where he recognized the key role AI plays in the financial field. During this time, his insights helped shape Imandra’s suite of financial products. Denis’ contribution to the computing logic of financial trading platforms includes several patents. He holds a master’s degree from the School of Computer Science and Finance from the London School of Economics.
Imandra is an AI-powered inference engine that uses neural subject AI to automate the verification and optimization of complex algorithms, especially in financial transactions and software systems. By combining symbolic reasoning with machine learning, security, compliance and efficiency can be improved, helping organizations reduce risks and increase transparency in AI-driven decision-making.
What inspired you and Dr. Grant Passmore to co-found Imandra, and how did your background influence the company’s vision?
After graduating from college, I made quantitative transactions and ended up in London. Grant pursued his PhD in Edinburgh and moved to Cambridge for the application of automatic logic reasoning to analyze the safety of autopilot systems (complex algorithms involving nonlinear computing). In my work, I also deal with many complex algorithms for nonlinear computing, and we realize there is a deep connection between these two fields. The way finance creates such algorithms is indeed problematic (highlighted by many news reports about “algorithms”), so we set out to change this problem by empowering engineers with automated logic tools to bring rigorous science and technology to software design and development. But what we ultimately create is indispensable to the industry.
Can you explain what neural symbolic AI is and how it differs from traditional AI methods?
The AI field has two areas: statistics (including LLMS) and symbols (also known as automatic reasoning). Statistical AI uses the information learned from the data it trains while recognizing patterns and performing translations. However, logical reasoning is terrible. Symbol AI is almost the same – it forces you to be very (mathematically) very precise with what you are trying to do, but it can use logic to reason in a logically consistent way and (2) no training data is required. The technology that combines these two regions of AI is called “neural symbols.” The famous application of this approach is DeepMind’s Alphafold project, which recently won the Nobel Prize.
Do you think Imandra stands out when leading the Neurosymbolic AI revolution?
There are many incredible symbolic reasoners (most in academia) targeting specific niches (such as protein folding), but Imandra gives developers an unprecedented automated analytical algorithm that has a larger application and a larger target audience than these tools.
How does Imandra’s automated reasoning eliminate common AI challenges such as hallucinations and improve trust in AI systems?
Through our approach, LLM is used to transform human requirements into formal logic and then analyze through a complete logical audit trail for inference engines. Although translation errors may occur using LLM, users are provided with logical instructions on how inputs are translated and can be verified by third-party open source software. Our ultimate goal is to bring viable transparency in which AI systems can interpret their reasoning in a logically independent way.
Imandra is used by Goldman Sachs and Darpa and others. Can you share a real example of how your technology solves a complex problem?
In our UBS Future Future of Finance competition No. 1, a good public example of Imandra’s real-world impact is highlighted (using the details of the Imandra code on our website). In a case study of encoding the regulatory documents they submitted to the SEC for UBS (UB), Imandra identified a fundamental and subtle flaw in the algorithmic description. The defect stems from subtle logical conditions that must be met that must be ranked in the order book, which humans cannot detect “manually”. The bank awarded us the first place (out of over 620 companies worldwide).
How your experience with Deutsche Bank has shaped Imandra’s application in the financial system, and what are the most influential use cases you’ve seen so far?
At Deutsche Bank, we process a lot of very complex codes that make automatic trading decisions based on various ML inputs, risk indicators, etc. As any bank, we must also comply with numerous regulations. What I realized is that this is very similar on a mathematical level to the research he did for autonomous driving safety.
Besides finance, which industries do you think have the potential to benefit the most from neuroaffirmation AI?
We’ve seen AlphaFold get the Nobel prize, so let’s definitely count that one… Ultimately, most applications of AI will greatly benefit by use of symbolic methods, but specifically, we’re working on the following agents that we will release soon: code analysis (translating source code into mathematical models), creating rigorous models from English-prose specifications, reasoning about SysML models (language used to describe systems in safety-critical industries) and business process automation.
Imandra’s regional decomposition is a novel feature. Can you explain how it works and its significance in solving complex problems?
The question that every engineer thinks of when writing software is “What edge situation?”. When their job is quality assurance, they need to write unit test cases, or they write code and consider whether the requirements are implemented correctly. Imandra brings scientifically rigorous questions to answer this question – it treats the code as a mathematical model and symbolically analyzes all its edge cases (while producing proofs about coverage integrity). This feature is based on a mathematical technique called “cylindrical algebraic decomposition” which we have “promoted” to the entire algorithm. It saves countless hours for our customers and discovers critical errors. Now, we bring this feature to engineers everywhere.
How does Imandra integrate with large language models, and what new features unlock for generating AI?
LLM and Imandra work together to formalize human input (whether it is source code, English prose, etc.), about its reasons, and then return the output in an easy-to-understand way. We use a proxy framework such as langgraph to coordinate this work and act as an agent that customers can use directly or integrate into their applications or agents. This symbiotic workflow solves many challenges of using AI tools that use only LLM and extends its applications beyond the training data you have seen previously.
What is your long-term vision for Imandra and how do you think of it changing the industry’s AI applications?
We believe that neural affirmation technology will pave the way for us to fulfill our commitment to AI. For most industrial applications of AI, symbolic technology is a missing component and we are excited to be at the forefront of AI’s next chapter.
Thank you for your excellent interview and readers who hope to learn more should visit Imandra.
