Onboarding personnel and training individuals can be a long and expensive process. A new developer in a bank needs to become familiar with documentation, guides, code bases, rules and best practices. By using an AI agent that automatically traverses the knowledge bases, collecting and indexing all the data, the user can get the answer to their specific question rather than something generic. A patient senior employee that is familiar with everything, is always available and never judges.
The answers are not only the most relevant, they’re tailored to a user’s way of thinking, in the right context and have a references to the source the answer was based on for verification.To minimise the exposure, the whole installation runs completely on-prem inside the bank’s data centres using a choice of top-of-the line open source LLMs.
Our pipeline is setup to track customer's existing sources of knowledge. We're parsing git repositories, Confluence online documentation and Google Drive changes. When someone makes a change, uploads a document or creates a branch it trigger a reindexing.
This means that newly added documents are available for inference the same day as they're added. As soon as you update your vacation policy, the AI will pick up on the new information and will start answering questions based on this new knowledge.
The whole system is deployable as a Kubernetes kustomize bundle and can run on-prem on your local data center or off-prem in any of the major cloud providers. This includes the AI APIs, Embedding models, Vector Database and all the integrations so your sensitive data never reaches the internet.
The system is also setup and affords different level privileges. Documents parsed can be segmented such that sensitive information isn't available to everyone - the AI API will feed the LLM only the information you are privy to. This way we have one system that can be used by the people manning the reception as well as the CFO without running the risk of leaking data.
Finally - we meet the customer where the customer is. A lot of the customers first experience will be a website, so that was a given. A developer who is in the middle of developing a feature doesn't want to stop, context switch and go use a web-app somewhere - they want their help right there in their code editor. So for them we built an IntelliJ plugin and a VSCode plugin.
Finally - we meet the customer where the customer is. A lot of the customers first experience will be a website, so that was a given. A developer who is in the middle of developing a feature doesn't want to stop, context switch and go use a web-app somewhere - they want their help right there in their code editor. So for them we built an IntelliJ plugin and a VSCode plugin.
A project lead in the middle of coordinating people on a project would love to have AI answer their questions right there in the group chat. For them we could build a Slack/MSTeams plugin. Regardless, the key idea is to be there, where the user needs you, and get them what they need without having to drop out of context or undue distraction.
