Why choose CompatibL AI
We are often impressed with clever responses of chat applications based on large language models (LLMs). However, a lot of research and engineering must be added before LLMs can become reliable and trusted components of enterprise trading and risk applications.
Based on our extensive research on LLM applications to quant finance presented at major industry conferences, CompatibL developed a sophisticated Co-Pilot that helps leverage the full power of AI technologies in performing time-consuming and tedious tasks in trading, risk management, and quant research.
To be clear, we do not propose that AI replaces humans in any trading or risk management task. Rather, our AI Co-Pilot works alongside humans to make their job less tedious and more rewarding.
This overview highlights two use cases—Model Governance and Document
Comprehension. There are many more and the list of what CompatibL AI can assist
with is continuously growing. CompatibL AI Co-Pilot software is free with a
minimal services engagement, and some of it is even open source.
Contact Sales to give it a spin.
Choose Between GPT and LLAMA2 Models
CompatibL AI offers a choice between the two leading LLM families: GPT models you can run in OpenAI or Azure cloud, and the open-source LLAMA2/CodeLlama models from Meta that you can run in our own private cloud or on-prem.
The partnership between Microsoft and OpenAI makes it possible to access GPT-4 and other OpenAI models inside your Azure Cloud.
If your firm already has data in Azure Cloud, the data submitted to GPT models will be subject to the same data protection policies.
The recent open-source release of the LLAMA2 model family by Meta and the widely anticipated successor models, combined with a vibrant open-source ecosystem around these models, provides an alternative to OpenAI that can be deployed on-premises or in a private cloud.
While OpenAI GPT model family remains a leader in all-around, general-purpose models for chat, rigorous testing performed by CompatibL indicates that the latest generation of LLAMA/CodeLllama for enterprise applications.
Model governance is a critical bank function with multiple stakeholders including the bank’s auditors and regulators.
Many model governance tasks require a tremendous amount of time and effort and can only be performed by a highly qualified team that consists of quants and risk experts. For example, each regulatory submission or internal model approval requires creating tens of documents that often reach 300 to 500 pages in length and manual review of tens and sometimes hundreds of thousands lines of code.
As a business function that involves dealing with extraordinarily large volumes of natural language text (model documents) and source code (model libraries), model governance is uniquely suited for the application of CompatibL AI.
CompatibL AI Co-Pilot uses LLMs to process and integrate information from model specifications, model test results, model revision history messages, existing documents, and regulatory guidelines.
It can look at every line of source code and every version control log message, perform an in-depth analysis of the prior documentation and release notes, and integrate all this data in a nuanced and sophisticated way. It can also cross reference the resulting documents with the specific lines of source code—something that is tremendously helpful to the bank examiners and internal risk control function, but rarely done because of the time and effort involved.
In addition to generating drafts of model governance documents, CompatibL Model Governance AI can also flag areas of concern inside the source code, including potential bugs, discrepancies between the source code of the model and how the model is described in the documentation, or the use of numerical methods and modelling techniques that were not intended or approved.
With CompatibL AI Co-Pilot, you will be able to:
For a critical business function that relies on rigorous analytics, securities trading involves a surprising amount of unstructured natural language documents. These documents include trade confirmations, term sheets, and offering memorandums for structured notes, derivatives, UCITs, CLOs, mortgages, and many others.
While most of these documents are generated based on templates, the wide variety of document formats and variations around them in the absence of a standard, widely accepted data model makes analysis of these documents for the purposes of data capture a monumental task. This task has resisted automation so far and is normally performed by many teams of highly qualified specialists across the firm.
CompatibL recognized that the emergence of LLMs for the first time offers an opportunity to automate comprehension of these complex documents, and developed sophisticated software that is customized and fine-tuned for each document category. When used as part of CompatibL AI Co-Pilot, this software can perform draft data capture for subsequent review by humans or validate the results of manual capture. In both cases, humans remain in control and sign off on the final result, while the amount of manual work is reduced, and data capture accuracy is improved.
To achieve these results, CompatibL used a combination of LLM customization standard techniques such as prompt engineering, RAG, and fine-tuning, as well as specialized methods such as logit processing to achieve reliable data capture in the prescribed format.