[ < ALL PROJECTS ]
WEBML/DATA

Medical Symptom Assistant

Medical Symptom Assistant is an educational health information tool that combines hybrid retrieval-augmented generation with safety-first urgency classification. It helps users explore possible conditions and follow-up questions using authoritative public health sources — not as a substitute for professional medical care.

FEATURES

  • Conversational Symptom Q&A

    Users describe symptoms in natural language and receive structured responses with possible conditions, follow-up questions, cited sources, and clear medical disclaimers.

  • Hybrid RAG Retrieval

    Combines pgvector cosine similarity and PostgreSQL full-text search in parallel, then merges results with Reciprocal Rank Fusion for more relevant medical context.

  • Urgency Classification

    A rule-based pre-retrieval classifier detects red-flag symptoms and vague queries so the assistant can prioritize urgent guidance or ask clarifying questions before generation.

  • Cross-Encoder Reranking

    Locally runs a cross-encoder model to rescore and reorder retrieved passages, improving the quality of context sent to the LLM.

  • Authoritative Data Sources

    Ingests and chunks content from MedlinePlus, WHO fact sheets, outbreak news, and Kaggle symptom-disease datasets into a searchable vector store.

  • Chat Sessions & Feedback

    Persists multi-turn conversations with session history and collects user feedback to support iterative improvements to safety and response quality.

TECH STACK

Next.jsReactTypeScriptFastAPIPostgreSQLpgvectorOpenAISentenceTransformersTailwind CSS

SCREENSHOTS