Cinchy

Fig 02 - Cinchy federated data collaboration

It is important to understand that the Cinchy platform does not take a one-off copy of original data and move it. Instead, bidirectional links are established with source systems such as ERP or CRM or other core systems. If data is changed in an underlying source system then that change is detected and reflected in Cinchy. If data is created or changed within Cinchy then that change is fed back into the appropriate source system. A detailed log of all changes is kept by the platform, showing exactly who or what process changed a data element and at what time. In this way the data mesh will not unravel. The Cinchy data is kept in a range of data storage options that the client already has perhaps a SQL Server or Postgres database.

This approach is novel but is not without challenges. With any federated approach to data management there is a tendency towards duplication of data, with different departments focusing on their own perspective. To be fair, Cinchy allows data ownership policies to be defined and delegated by data owners, which should minimise this issue. Similarly, data quality remains a concern, but that is not unique to Cinchy. Indeed, with a group of people dedicated to collaborating on the data, data quality improvements are likely to be made and indeed propagated back to the source systems. Cinchy has an interactive knowledge graph that shows the relationships between data products and can be used to find data that is needed by (authorised) business users.

Cinchy makes use of Open AI’s ChatGPT in several ways, including searching the network of data within Cinchy to help users find what they need. Indeed, this suggests an interesting possible use case for Cinchy. At present most large companies have banned the use of public generative AI products at work due to security concerns. Instead, many companies have taken large language models such as Meta’s LLaMA and trained them on specific datasets such as customer service records or supply chain data. A limitation of this is that a very specific dataset may limit the ability of the AI model. Cinchy’s ability to allow collaboration across many different data products make it a potential source for company-specific AI model training that is not restricted to specific applications and would grow in value over time as more data products are created. Cinchy can also be configured with alternate large language models instead of ChatGPT, and also more traditional AI/ML technologies if desired.

As the core Cinchy platform matures the company is likely to develop application-specific packages on top of the core platform, perhaps for particular vertical industries. One consequence of such a data collaboration approach is that it can potentially reduce the complexity of data integration.  It allows the extension of a data mesh approach to operational data and not just to data used for analytic purposes.

The bottom line

Cinchy offers an innovative approach to data collaboration, and fits very well within the modern trends towards data mesh architecture, which assumes a decentralised, federated approach to data ownership.