What is psycho?

Psycho Python is a backend system for a campus mental health support chatbot, built with FastAPI. Students log in and chat with the system, and their messages stream back word by word rather than appearing all at once. Students and administrators are kept separate through basic login roles, and administrators get access to reports that students never see. Behind the scenes, each message passes through a set of specialized agents built with LangGraph, a framework for coordinating multiple AI steps. A supervisor first decides whether the message is ordinary conversation, a request for counseling, or a sign of risk, and only counseling or risk cases trigger a search through a knowledge base. That search combines a vector database called Chroma with a keyword-based search method, then reranks the combined results locally. If LangGraph, Chroma, or the OpenAI embedding service is unavailable, the system automatically falls back to simpler keyword search so a demo does not break. Messages flagged as high risk are checked against a dictionary of risk terms as well as evaluated by the language model itself. When risk is confirmed, the system writes a report recording emotional tags, a risk score, and a summary, saves it to a MySQL database, and can send an email alert and log the case to an Excel file, all handled through background jobs so the student's chat is not delayed. Short-term conversation memory is kept in Redis so recent context does not need to be reloaded from the database every time. The project can run against a local, fine-tuned language model served through Ollama, or against any OpenAI-compatible API, and it ships with Docker Compose configuration to start MySQL, Redis, and the application together. It also includes scripts for evaluating the retrieval system using standard measures such as Recall, Precision, and NDCG. The README is written primarily in Chinese and documents extensive environment variables for configuring the knowledge base, database connections, and alerting behavior.