Semantic Framework for Transactional Applications
Yes, the title is a mouthful. But, it needed to be said.
As a data engineer, one often wonders about how to leapfrog my applications to the current gen. Application architectures have evolved quite quickly in the past 3–5 years, while our applications in a corporate enterprise setting, haven’t.
So, how do I bring my applications to leverage AI and the power of semantics? Allow users to interact with your transactional data stores, like they were knowledge bases?
Perhaps, you can vectorize your transactions.
Identify the lowest level business friendly transaction you can represent via an embedding. An embedding, is a vectorized representation of your transaction in english.
As an example, if your table has 3 fields: first name, last name, date of birth, it could be represented as “first name, last name, 01/25/2025”, which is a perfectly simple transaction to convert to an embedding.
Let their embeddings define the semantic similarity via cosine similarity or any of other options available for vector databases.
