| # Medicode | |
| ## Getting started | |
| ### Prerequisites | |
| 1. ffmpeg for audio processing in Bumblebee's speech-to-text serving: `brew install ffmpeg`. | |
| 2. Postgres and pgvector for storing data and vector embeddings: `brew install pgvector`. | |
| ### Running the server | |
| To start your Phoenix server: | |
| - Run `mix setup` to install and setup dependencies | |
| - Run `mix build_code_vectors` to download the ICD-9 codelist, precompute vectors, and store the results in the database. | |
| - Start Phoenix endpoint with `mix phx.server` or inside IEx with `iex -S mix phx.server` | |
| Now you can visit [`localhost:4000`](http://localhost:4000) from your browser. | |
| Ready to run in production? Please [check our deployment guides](https://hexdocs.pm/phoenix/deployment.html). | |
| ## Deployment | |
| The app is configured to deploy to Fly.io via a `fly.toml` file. To deploy, run `fly deploy` within the app's directory. | |
| ### Precomputing code vectors | |
| To build the code vectors for the ICD-9 codelist for the deployed environment: | |
| 1. Connect to the server with `fly ssh console`. | |
| 2. Run `/app/bin/medical_transcription eval Medicode.Release.precompute_code_vectors`. This will prepare the vectors in the database if they are not present. | |
| ### Livebook | |
| In addition to connecting to the deployed application via `iex`, Livebook supports connecting to the running application. Connecting a Livebook instance to the deployed application involves the following: | |
| 1. Install and setup Wireguard with a peer connection for Fly.io: [Step by Step](https://fly.io/docs/networking/private-networking/#install-your-wireguard-app) | |
| 2. Install and start Livebook: [Livebook.dev](https://livebook.dev/) | |
| 3. [Connecting Livebook to a Production App](https://fly.io/docs/elixir/advanced-guides/connect-livebook-to-your-app/) requires a node name and cookie value: | |
| - Node name: `medical-transcription-cpu@myipfromfly` ("myipfromfly" can be retrieved with `fly ips private --app medical-transcription-cpu`) | |
| - Cookie value: `0gfxcPtwryKxI2O1N0eFAg9p4MJGC-oUGShgj_wgvNEGiba5EDEJFA==` (this value is set in `fly.toml`) | |
| ## Run in Docker | |
| 1. Create a local volume: `docker volume create ml-data` | |
| 2. Ensure the volume is writeable: `docker run --rm -v ml-data:/data busybox /bin/sh -c 'touch /data/.initialized && chmod 1777 /data'`. More background: | |
| - What the initial `1` for `chmod` means: <https://www.linuxnix.com/sticky-bit-set-linux/> | |
| - How to update the permissions of a mounted volume: <https://serverfault.com/a/984599> | |
| 3. Build the image with: `docker build . -t headwayio/medical_transcription` | |
| 4. Run a container with: `docker run --env-file ./.env -p 4000:4000 headwayio/medical_transcription` | |
| ### Caveats | |
| You may need to make a few changes to get the app running in Docker at the moment: | |
| - In `lib/medical_transcription/application.ex`, comment out the `DNSCluster` child spec. | |
| - In `rel/env.sh.eex`, comment out the `ERL_AFLAGS`, `RELEASE_DISTRIBUTION`, and `RELEASE_NODE` environment variables. | |
| ## Learn more | |
| - Official website: <https://www.phoenixframework.org/> | |
| - Guides: <https://hexdocs.pm/phoenix/overview.html> | |
| - Docs: <https://hexdocs.pm/phoenix> | |
| - Forum: <https://elixirforum.com/c/phoenix-forum> | |
| - Source: <https://github.com/phoenixframework/phoenix> | |