Basically, you can upload documents and then you have access to all the information of your documents at your fingertips.
There is a Q&A tool to answer questions across your entire knowledge base and an AI assistant built-in, that will only rely on your knowledge base for its facts. Think of it like a Bing Chat but that will search the documents you have in your knowledge base instead of the web.
Tech details: To build the knowledge base, we use a combination of embeddings, summarizing and pre-processing to allow the ability to have full context of what is in a given document (say 40+ pages) and conditional prompting depending what the document is (ex: an excel file will be processed a bit differently than a word/pdf file). For the AI assistant, it’s a mix a clever prompting, using the Q&A tool when in need of information and “self-reinforcement” / “grounding” to keep it on the rails.
In the future, I want to give more tools to the assistant (improved math, task automation, internet access) and an API to make it easy to add/update your knowledge base or integrate knowledge-augmented models into products easily. Also, with LLMs having their Stable Diffusion moment, I think in the next few months we’ll be able to have a self-host version of the product which will be very exciting.
I’d love to get some feedback (on the features, pricing, use cases, etc.) or answer any questions you may have!
I wanted to show you Alceos, an interactive knowledge base that makes it easy to get AI to work with your data.
Demo: https://youtu.be/EcjxFN_SWbg
Basically, you can upload documents and then you have access to all the information of your documents at your fingertips.
There is a Q&A tool to answer questions across your entire knowledge base and an AI assistant built-in, that will only rely on your knowledge base for its facts. Think of it like a Bing Chat but that will search the documents you have in your knowledge base instead of the web.
Tech details: To build the knowledge base, we use a combination of embeddings, summarizing and pre-processing to allow the ability to have full context of what is in a given document (say 40+ pages) and conditional prompting depending what the document is (ex: an excel file will be processed a bit differently than a word/pdf file). For the AI assistant, it’s a mix a clever prompting, using the Q&A tool when in need of information and “self-reinforcement” / “grounding” to keep it on the rails.
In the future, I want to give more tools to the assistant (improved math, task automation, internet access) and an API to make it easy to add/update your knowledge base or integrate knowledge-augmented models into products easily. Also, with LLMs having their Stable Diffusion moment, I think in the next few months we’ll be able to have a self-host version of the product which will be very exciting.
I’d love to get some feedback (on the features, pricing, use cases, etc.) or answer any questions you may have!
Cheers,
Iliane