In today’s highly regulated world, ensuring compliance to the ever changing regulatory frameworks is a challenge for most businesses. The audit and compliance departments often find themselves lost in a pile of rules, regulations, acts, directions. The frequency of notifications and circulars which amend the existing regulations is quite high. To make matters worse, there are multiple regulatory agencies that need to be complied with.
The past decade saw the rise of “RegTech” software. An industry exists around expert consultancy services to handle regulatory changes, including the likes of Thomson Reuters, Price Waterhouse Coopers, Deloitte, Ernst & Young et al. However, the RegTech software solutions were merely aggregators of multiple sources of information. The aggregation of information into piles of documents increases cognitive load of the audit and compliance teams. The search functionality in these systems is the similarity search based on keywords and frequencies of words.
However, success of generative AI and foundation models opens up new possibilities for regulatory documents. This is because for unstructured data in regulatory text documents the large language models (and multimodal models) excel at the following
- Comprehension of the textual content
- Extraction of structured information
- Reasoning and deductions
- Understanding image content in charts, figures and tables
- Summarization of textual content
- Transformation from one form to another
- Verification of contextual information
As a team which specializes in artificial intelligence, at dataeaze systems we saw the opportunity to address the pain points of compliances and audits. So, we have built a solution “Complieaze”, which helps reduce the workload of the compliance officers. The application is powered by RegLLM, a domain adapted, instruction fine tuned and preference aligned large language model.
Features
Technology
- Small language models
-
-
- Small models fine tuned to excel on the use cases of regulatory compliance
-
- Dataset curation
-
-
- Training data for RegLLM is from public sources and augmented by synthetic data generation techniques
-
- Knowledge Graph
-
-
- LLM is used to build a knowledge graph representing the change relationship between documents
-
- Agentic architecture
-
-
- Simple natural language interface to retrieval augmented generation (RAG) based Q & A as well as tailored functionalities is achieved through agents architecture
-
- Extensibility
-
-
- Solution is extensible to new industry verticals, new feature requests, deployment platforms, geographies
-
- Platform Support
-
-
- The solution is available on all cloud platforms, on-premise and hybrid platforms
-
- Cost Effective
-
-
- The solution has been designed considering the cost sensitive nature of Indian enterprises
-
- Ease-of-use
-
- The solution can be used as a web application, API and also model weights
The solution has moved to the production stage and been deployed at Aditya Birla Group. It has been part of AWS Gen AI catalyst programme. It is under deployment on Azure Marketplace as well.
Looking Forward
As artificial intelligence specialists we have built a solution to reduce time spent in reading documents based on our insights. But we are open to suggestions and feedback from all of you.
We have an open weights policy. We are releasing our model weights on huggingface https://huggingface.co/dataeaze
We are working on model improvements and new features like risk control matrix, audit evidence, audit sampling, verification of audit evidence, audio input and output.
Although we started with the BFSI sector with focus on audit, the framework is extensible to almost any business domain where regulatory compliance is required.
Demo
For a quick live demo visit :
or check out videos at
https://www.youtube.com/playlist?list=PLj6nyFXvFsHY33qCJ8XL4BPFsI9iOJ1Vd
and you could reach us at contactus@dataeaze.io
