hamzaazizai/medical-chatbot-web repository overview

⁠Mr. Healthcare

A rule-based medical question-answering chatbot (a B.Tech project). It answers basic questions about 658 diseases and their symptoms, causes, prevention, treatments, complications and descriptions from a Neo4j knowledge graph, and says plainly when it cannot answer. Every answer is a fact from the graph; nothing is generated.

This is a student prototype for local use. It is not medical advice and is not built for public deployment.

Source, documentation and releases: github.com/hamza-aziz-ai/Medical-ChatBot⁠

⁠Images

Both images are needed, together with Neo4j 5.26, which the release's compose.yaml starts for you.

⁠Tags

• One version tag per release, for example 1.1.13; see the Tags tab. There is no latest: pull an explicit version.
• Version tags are immutable and never move.
• Each tag is copied byte for byte from ghcr.io/hamza-aziz-ai/medical-chatbot-api and -web after the release passed its tests, evaluation and an end-to-end smoke test, so the digests are identical on both registries.
• linux/amd64. Each image carries an SBOM and SLSA provenance: docker buildx imagetools inspect hamzaazizai/medical-chatbot-api:<version> --format '{{ json .Provenance }}'.

⁠Run it

You need Docker with Compose.

1. From the GitHub release⁠ of the version you want, download compose.yaml and env.example. Save env.example as .env next to compose.yaml and set NEO4J_PASSWORD.

2. Point Compose at the Docker Hub images (PowerShell: $env:API_IMAGE = "..."):

   export API_IMAGE=hamzaazizai/medical-chatbot-api:<version>
   export WEB_IMAGE=hamzaazizai/medical-chatbot-web:<version>
   export LLM_UNDERSTANDING=off
   

   Copy

3. Start the database, import the graph once, then start everything:

   docker compose up -d --wait neo4j
   docker compose --profile init run --rm importer
   docker compose up -d
   

   Copy

4. Open http://127.0.0.1:8080⁠ and ask, for example, "what are the symptoms of asthma".

Only the web port is published, on 127.0.0.1 (WEB_PORT changes it). LLM_UNDERSTANDING=off keeps every question on your machine and uses the rules alone; with it on, the API asks a local Ollama at host.docker.internal:11434 to help read questions and falls back to the rules whenever that is unavailable.

⁠Privacy and limits

• Questions travel in the URL, so the API marks every response Cache-Control: no-store, and nginx keeps no access or error log. The API logs never contain questions.
• The API is rate-limited in nginx and accepts only short GET questions. It has no user accounts: do not expose it beyond your own machine without adding authentication and HTTPS.
• The bundled dataset's redistribution rights are unverified.