A lightweight cohort discovery application that combines:

• MedCAT annotations API (clinical concept extraction)
• NL2DSL API (natural language → JSON cohort query via Ollama)
• WebAPP (frontend + backend for the full Cohorter experience)

Live demo: https://cogstackcohort.sites.er.kcl.ac.uk/

• NL2DSL/
  API service that accepts natural language queries, calls Ollama to translate them into a JSON DSL query, and uses MedCAT to normalize medical terms.

• WebAPP/
  Full application (frontend + backend) providing the UI and orchestration of the workflow.

• Docker Desktop (or Docker Engine) with Docker Compose
• (Optional) GPU setup if you want Ollama GPU acceleration (depends on your host OS)

From the repo root:

docker compose up --build

Or run in background:

docker compose up --build -d

• Web UI: http://localhost:3000
• MedCAT API: http://localhost:5000
• NL2DSL API: http://localhost:3002
• Ollama: http://localhost:11434

NL2DSL uses:

• OLLAMA_URL (default in compose: http://ollama:11434/api/generate)
• OLLAMA_MODEL (default: gpt-oss:20b)
• MEDCAT_URL (default: http://cohorter-medcat:5000)
• ALLOW_ORIGINS (default: *)

WebAPP uses:

• NL2DSL_URL (default: http://cohorter-nl2dsl:3002)

The WebAPP/Dockerfile supports a build arg random to generate random data:

WEBAPP_RANDOM=true docker compose up --build

(If you don’t need random data generation, keep it false.)

You can work on each component independently:

• NL2DSL/ — API service (see its folder README)
• WebAPP/ — Node app (server + client)

Each folder contains its own Dockerfile and scripts.

• MedCAT model packs and any potentially sensitive data must not be committed to the repository.
• Ollama models are stored in a Docker named volume by default (ollama: in docker-compose.yml) so they persist across restarts.

docker compose down
docker compose up --build

This deletes the named volume that stores Ollama models:

docker compose down -v

docker compose logs -f

Built within the CogStack ecosystem to support clinical cohort discovery workflows.