LangDoc (LangDoc.AI) is an accessible symptom checker and anamnesis tool that can interview patients about their symptoms in natural language via a simple chat interface - in all prominent languages.
⇒ Interviews patients in natural language via chat in an empathic and professional manner
⇒ Supports all large languages out of the box
⇒ Dynamically gathers information on relevant symptoms
⇒ Compiles a comprehensive overview of symptoms, patient history, and other patient data
⇒ Generates a concise summary of the case to inform the clinician's anamnesis subsequently
⇒ Planned features: voice chat and direct interface with hospital information systems (e.g. FHIR)
LangDoc in action
LangDoc’s use cases in hospitals and practices are apparent and manifold: For one, it can handle efficient and accessible pre-screening, registration, and anamnesis of patients to speed up patient intake. It breaks down language barriers which are common in larger hospitals.
On the backend, LangDoc runs on a LangChain natural language computing architecture in an LLM-agnostic way, with GPT-3.5 and GPT-4 as current foundation models. The frontend is currently realized via a simple Discord chatbot interface. Both frontend interface and backend are deployed via FastAPI on Railway.
This stack has been optimized for prototyping purposes, later iterations will include a use-case-specific front-end (e.g. HIS interface via FHIR, end-user web apps, etc.) and local, open-source foundation models (e.g. Llama) for data security reasons.
LangDoc is being built by me, Tim Farkas. I am a medical student at Charité Berlin, a digitalisation / AI enthusiast, and I love to build things! See my LinkedIn for other things I built in the past.
Originally, LangDoc was designed to be a diagnostic tool to gather patient information and dynamically rank the likelihood of different diagnoses. It can still do this, and quite well.
diagnostic capabilities of LangDoc