.png)
myAuxiliumis the result of 2 years of intensive R&D
by our expert data scientists inmachine learning.
We use the most advanced technologies – continuously improved – to overcome the limitations of public large language models like ChatGPT.
Ourmission is to provide you with a precise, reliable, and secure tool thatmeets your professional requirements and helps you efficiently accomplish yourmost time‑consuming tasks every day.
Vectorizationin artificial intelligence involves transforming files – such as text documentsor images – into vectors.
A vector is essentially a list of numbers representing the key characteristics of the data, which allows machines to understand the content and perform mathematical operations on it.
Retrieval-Augmented Generation (RAG) is an AI technique that combines generative models with information retrieval. Before generating an answer, the model first retrieves relevant information from an external knowledge base – for example, your own data.
Instead of retraining the AI model with new information, RAG augments the model’s knowledge by feeding in up-to-date external data, ensuring the answers are based on the most current information.
By using your external data, RAG greatly reduces the risk of hallucinations, where the model might otherwise invent incorrect facts.
In myAuxilium, the use of RAG combined with our other technologies drives the risk of AI hallucination almost down to zero.
Tree of Thoughts (ToT) is a conceptual framework designed to enhance the reasoning capabilities of large language models. This approach mimics human problem-solving strategies by allowing the model to explore multiple potential solutions in a structured way – much like branches of a tree.
ToT uses search algorithms to traverse different “branches” of thought and evaluate intermediate steps, improving the quality of the final answer. In myAuxilium, this means your AI assistant performs an in-depth analysis of your query and evaluates the response thoroughly to ensure it’s accurate and comprehensive.
By continuously evaluating intermediate reasoning steps, the Tree of Thoughts approach also helps reduce errors and hallucinations in the output.
myAuxiliumensures that all of your uploaded files are fully vectorized, providinga rich knowledge base tailored to your data with no compromises.
Foreach question you ask, the system mobilizes multiple AI agents inparallel, allowing it to dive deeper into every aspect of your query.
Your question is broken down into different logical components, which are analyzed separately by these AI agents. Each agent searches the entire vectorized corpus for patterns matching its component of the query. They then reunite after completing their searches, combining their findings to create as many cognitive connections as possible between the different parts of your question. This neural search mechanism ensures nothing is overlooked and you get the most precise answer from your data
The Langchain Conversational Memory module (part of the Langchain framework)manages the storage and retrieval of conversation history.
It allows the AI assistants to remember the context of past interactions and use that information to generate relevant, coherent responses as the conversation continues. In practice, this means you can have long, complex dialogues with myAuxilium, and it will maintain context and understanding throughout.
Alldata you handle in myAuxilium remains exclusively on our secure serversin a sovereign hosting environment (OVH). Once your files are vectorized, theoriginal source files are immediately deleted from our system.
Whenour AI assistants process your vectorized data, they interact with OpenAI’s APIto generate responses. However, none of your data is used to train OpenAI’smodels, and OpenAI automatically deletes any conversation data after 30days.
Additionally, all interactions with myAuxilium are protected by HTTPS encryption, ensuring your exchanges are secure.
Your files are pre-processed before the vectorization step to optimize results. This process includes:
• Chunking : Splitting textual data into smaller segments for analysis.
• Summarization: Generating a concise summary of each segment via an API (OpenAI).
• Vectorization: Converting each summary into a vector and building a vector database that links these vectors to the corresponding segments of the original content.
After this indexing process, the original files are removed from the server.
Likewise, if you delete documents from a collection or delete an entire collection, those documents’ data are removed from the vector database.
However, myAuxilium maintains comprehensive logs that allow us to trace which documents were used in retrieval (RAG) processes, even if the vector data for those documents has been deleted. This means you have full traceability for legal compliance or oversight purposes.