CogniLoop

Lazy cat OIDC login has been integrated. CogniLoop is an AI learning system for self-study, test preparation and knowledge accumulation. You can organize course materials, documents, notes and test papers into a knowledge base, and let the system conduct questions and answers, questions, corrections, study portrait analysis and weakness tracking based on these materials. ##Who is suitable for use - Individual users who are preparing for exams, brushing questions, or reviewing courses. - Learners who need to organize PDFs, handouts, and notes into a searchable knowledge base. - People who want to automatically generate exercise questions based on their knowledge. - A group that needs to form a learning circle with classmates and friends to set questions, challenge each other and share information. ##First use 1. Open the app and click "Log in using lazy cat account". 2. After logging in, enter the dashboard. 3. If you are an administrator, go to "System Management" and check the LLM, Embedding, and OCR configurations. 4. After confirming that the model configuration is available, create a knowledge base and upload the data. 5. After the document is parsed, you can use the knowledge base question and answer, intelligent question setting and learning portrait functions. ##Configure LLM and Embedding Before using it for the first time, it is recommended to enter the "System Management" configuration model service. There are two most important types of models in CogniLoop: the LLM core model and the Embedding vector model. They have different uses and cannot simply replace each other. ### LLM Core Model LLM is a chat model responsible for "understanding, generation and reasoning." It is used for: - Generate answers based on the content of the knowledge base. - Generate practice questions and test papers. - Correct subjective questions. - Analyze wrong questions and weak knowledge points. - Generate learning portraits and review suggestions. Configuration location: System Management-> System Settings-> LLM Core Model. Need to fill in: - API Key: The key of the model service. - Base URL: OpenAI-compatible interface address, such as `https://api.openai.com/v1`, or the address of the third-party compatible interface you use. - Model Name: Chat model name, such as `gpt-4o-mini`,`gpt-4o`,`deepseek-v4-flash`, etc. If you use a chat model like 'deepseek-v4-flash', you can configure it in the LLM core model. As long as the service provider provides an OpenAI-compatible interface, it can be used for question setting, correction, question and answer generation, and learning portrait analysis. ### Embedding vector model Embedding is the model responsible for "turning text into vectors." It is not responsible for chatting and generating answers, but is responsible for knowledge base retrieval. The system will cut the uploaded document into fragments, and then use the Embedding model to convert it into vectors and store it in the database. Later, when the user asks or sets questions, the system can find relevant information from the knowledge base. Embedding is used for: - Vectorized indexing of documents after uploading. - Relevant segment retrieval in knowledge base Q & A. - Set questions based on the content of the knowledge base. - RAG retrieval enhances generation. Configuration location: System Management-> System Settings-> AI Service Configuration-> EMBEDDING vector. Need to fill in: - API Key: Embedding service key. It can use the same service provider as LLM or it can be different. - Base URL: The OpenAI-compatible interface address of the Embedding service. - Model Name: A special vector model name, such as `text-embedding-3-small`,`bge-m3`,`qwen3-embedding`, etc. Note: The chat model cannot usually be used as an Embedding model. For example,`deepseek-v4-flash` can be used as the core model of LLM, but it cannot replace Embedding. To use the knowledge base Q & A and document-based question setting, an additional configuration must be provided for the available Embedding model. ###Configuration recommendations Minimum available configuration: - LLM core model: Must be configured. Otherwise, question setting, correction, question and answer generation and learning portraits will not be available. - Embedding vector: Configuration is highly recommended. Otherwise, the knowledge base index cannot be established normally after uploading a document, and the knowledge base questions and answers and questions based on the document will be affected. - OCR recognition: Optional. Only if you need to scan test paper pictures or PDFs to generate test paper templates. After the configuration is complete, click "Test Connection" first. After the Embedding test is successful, if a document has been uploaded, it is recommended to click "Rebuild All Index" to allow the old document to be indexed again using the current vector model.