Leveraging the change brought about by generative AI technologies to augment existing and build new products/services by leveraging your unique data assets.
The world of AI is at a historical inflection point with the emergence of Generative AI capabilities. While products like ChatGPT have attracted significant attention it is the expansion of these models with data that creates true use cases for new products and services. Recent advancements in ML techniques have been coupled with the scale of Cloud and proliferation of data.
By leveraging Cloud services to train and run new Large Language Models (LLMs) companies can extract value from their data like never before. The value of generative AI applications will be driven by data, and applications and companies with sufficient data gravity and the ability to extract value from it via customising LLMs will be disruptors across all industries. This requires the scale of Cloud and Serverless, as well as careful design and operation to protect data, mitigate hallucinations and ensure compliance.
To gain value from general LLMs they need to be augmented with custom proprietary data sources. This requires knowing what data the organisation has, where it is stored and how to access and surface it to the LLM. In addition data compliance categorisation, anonymisation (e.g. vectorisation) and PII protection concerns need to be highlighted and mitigated.
Data sources need to be aggregated and orchestrated at scale. Serverless Data Lakes and ETL process can be leveraged to extract data from heterogeneous locations and formats, enabling access to LLM training. The scale of data to generate completely new LLMs is typically out of the range of any one company's data set, yet customisation of an existing LLM requires a much smaller data volume. Existing generic LLM models (e.g. Amazon Titan) can act as Foundational models and adapted with custom data.
Model Hosting and Observability
Once the foundational LLM has been adapted with custom data an API interface is built around the inference engine to provide access to the model to applications and services. This API can be built for a range of internal products as well as productised to third parties. In the fast moving world of regulation around LLMs strong observability and safeguards need to be built into this interface for audit and prevention of data compliance issues.
No one likes to be the last to the party. If you’re looking to get into serverless, we’ve got experts waiting to help.
Fill in your details below, or simply give us a call on +442045716968