MedChat: A Fully Offline Multimodal AI System for Clinical Anamnesis

MedChat is a locally deployable virtual physician framework integrating an LLM-based medical chatbot with a diffusion-driven avatar for automated and structured anamnesis. The system was fine-tuned using a corpus of LLM-generated medical dialogues derived from publicly available symptom-disease datasets. The dataset was uploaded by Jan Benedikt Ruhland on April 16, 2026.

Use Cases

• Training or benchmarking privacy-preserving medical chatbots based on the described LLM-generated dialogue corpus.
• Developing multimodal clinical interfaces that integrate text and avatar components based on the described conditional diffusion model.
• Evaluating the performance of locally deployable AI systems for structured anamnesis based on the reported latency and coherence metrics.
• Researching secure database interfaces for isolating patient data from model inference processes as described in the framework.

Strengths

• The system is reported to operate fully offline on consumer-grade hardware with an average latency of 2.9 ± 0.3 seconds.
• The training corpus was derived from publicly available symptom-disease datasets, enabling scalable and privacy-preserving training.
• The framework implements a secure and isolated database interface to ensure complete separation between patient data and model inference.

Limitations

• The dataset is a 1.1 MB DOCX file; the actual data content, column definitions, and row count are unknown and require inspection after download.
• Clinical validation is described as ongoing, indicating the system's medical efficacy may not yet be fully established.
• The description metadata is limited; data quality and structure must be manually verified.

Provenance

Source

Jan Benedikt Ruhland via figshare

Collection Method

The chatbot was fine-tuned using a corpus of LLM-generated medical dialogues derived from publicly available symptom-disease datasets.

Time Range

The dataset itself does not specify a temporal coverage for the underlying data.

Freshness

Last updated 2026-04-16 05:17:56

Geography

The dataset does not specify a geographic coverage.