Member-only story
From LangChain to LangGraph: Making the Multi-Model DrugBot Personal and Teachable
Adding memory and learning ability to the human-in-the-loop chatbot

Drug trials test new medications for safety, effectiveness, and efficacy in humans. These critical studies are essential for developing and approving life-saving therapies. While drug trials offer hope to countless patients with serious illnesses, many remain unaware of their eligibility or the potential benefits. A user-friendly drug trial information system can bridge this gap. It should contain a database with authoritative information and an easy-to-use frontend, with which patients can navigate the complex world of clinical research and discover beneficial opportunities.
In my previous articles, DuckDB as a DrugDB: a Free and Simple Multi-Model Drug and Trial Database and Tailor a Multi-Model Chatbot for a Multi-Model DuckDB, I built a drug database called DrugDB and its corresponding chatbot named DrugBot. DrugDB is built on top of DuckDB, a multi-model database. Users can combine SQL, graph, vector, and full-text queries to explore a network of drugs, disorders, mechanisms of action (MOA), and drug trials. However, its technical nature can be daunting for average patients. To bridge this gap, I developed a user-friendly chatbot that leverages LangChain and…