Constructor Research product update: July

Knowledge models

Constructor Knowledge models is a technology for managing and interacting with users’ knowledge. 

Knowledge models allow users to leverage Large Language Models (LLMs) to interact with their own documents and data using natural language. This allows user to get verified answers and insights.

Users can create their own Knowledge models, upload documents, and start chatting with AI over their content. For each response generated, a Knowledge model provides a set of links to the documents used. This allows users to verify all responses.

Users can create multiple persistent chat sessions with a Knowledge model. This enables users to return to any chat at any time and continue the conversation from where thet left off. 

The following file formats are supported for uploading, processing, and interaction: .txt, .pdf, .ppt, .pptx, .ipynb, .json, .md, .py, and .csv. More formats will be added soon.

Document management in Knowledge models

Users can view lists of documents uploaded to their Knowledge model. The processing status is displayed for each document: awaiting processing, being processed, ready for use, or encountering an error during processing. Text documents can also be viewed after uploading. These updates improve content navigation and status tracking in knowledge models.

Knowledge model chat improvements

Chats now have infinite scrolling, which allows users to browse their entire conversation history. Each response now includes an option to provide feedback on its quality to aid in the improvement of the Knowledge model.

Folder upload to work desks and Knowledge models

Users can upload folders to their desks and Knowledge models. This simplifies the process of uploading multiple files or entire folder structures at once.

PDF viewer in VS Code environment

Users can view PDF documents directly in the VS Code environment, which makes it easier to identify the required information.

Publishing Knowledge models and configuring access

Users can publish their Knowledge models and share them with others so they can also interact with them. When publishing, the owner can set the required access level: Public, Confidential, Secret, or Top Secret. Only users with the specified access level can access and interact with the published knowledge model. This feature allows users to share Knowledge models while maintaining required security levels and protecting sensitive information.

Enabling or disabling documents in Knowledge models

Users can disable individual uploaded documents in their Knowledge models to prevent them from being used for response generation. This feature allows users to quickly refine their Knowledge models by disabling documents containing obsolete or irrelevant information.

Viewing PDF and Markdown documents in Knowledge models

Users can view the content of PDF and Markdown documents uploaded to their Knowledge models. This feature enables users to easily identify documents and find the information they need.

Configuring document access in Knowledge models

Users can set access levels for each document uploaded to their Knowledge models, similar to the Knowledge model itself: Public, Confidential, Secret, or Top Secret. When users interact with the published Knowledge model, they receive responses based only on documents that correspond to their access level. This feature allows Knowledge model owners to implement more precise access control over content in their Knowledge model.

Knowledge Model Explorer

Explorer is an extension for Knowledge models designed to assist scientists in exploring new research areas. Through chat interaction with the user, Explorer gathers necessary information about potential research topics. It explores the OpenAlex database of published research papers to understand the current state-of-the-art in selected fields. Explorer then generates several formalisms that can serve as research topics. Once a formalism is chosen, Explorer generates hypotheses and research plans to accelerate the ideation process for new research projects.

Knowledge Browser

Knowledge Browser allows the owner of a Knowledge model to explore the taxonomy of its knowledge. Based on uploaded documents, it extracts a set of terms and concepts, enhances some through the WikiData ontology, and presents them as a two-level mindmap. Each node corresponds to a key term in the model and includes references to documents using that term. The Knowledge Browser helps users to explore and visualize the knowledge extracted from their documents and better understand the possible gaps in this knowledge.

Enhanced document references in chat

Knowledge model responses in the chat include direct references to the specific parts of documents that were used to generate each response. The reference list has been expanded to include the referenced part of the document. This allows users to better understand where each specific bit of response is built upon and adds additional verification to the knowledge model responses.

Improved file uploading

Users can upload folders with more than 50 files. This improvement makes uploading files and documents much more convenient.