refactor: Move CodeSearcher and CodeVector into Coding context

lib/medical_transcription/code_searcher.ex

@@ -1,20 +0,0 @@
-defmodule MedicalTranscription.CodeSearcher do
-  @moduledoc """
-  Takes a portion of text and searches for closely matching results within a list of vectors, by using AudioTagger's
-  SemanticSearch module.
-  """
-
-  alias MedicalTranscription.{CodeVector, Utilities}
-
-  def process_chunk(text, opts \\ []) do
-    k = Keyword.get(opts, :num_results, 5)
-
-    search_vector_for_db = Utilities.compute_vector_as_list(text)
-
-    CodeVector.find_similar(search_vector_for_db, k)
-
-    # -- Remove matches that don't exceed a given threshold
-    # TODO: This depends on receiving a similarity score from the Postgres query
-    # |> Enum.filter(fn {_index, score} -> score >= similarity_threshold end)
-  end
-end

lib/medical_transcription/{code_vector.ex → coding.ex}

@@ -1,28 +1,19 @@
-defmodule MedicalTranscription.CodeVector do
+defmodule MedicalTranscription.Coding do
   @moduledoc """
-  Represents a code and its description, along with a vector embedding for its description.
+  Takes a portion of text and searches for closely matching results within a list of vectors, by using AudioTagger's
+  SemanticSearch module.
   """
-  use Ecto.Schema
-  alias Ecto.Changeset
-  alias MedicalTranscription.Repo
 
+  alias MedicalTranscription.Repo
   import Ecto.Query
   import Pgvector.Ecto.Query
 
-  schema "code_vectors" do
-    field :code, :string
-    field :description, :string
-    field :description_vector, Pgvector.Ecto.Vector
-  end
-
-  def changeset(code_vector, params \\ %{}) do
-    code_vector
-    |> Changeset.cast(params, [:code, :description, :description_vector])
-    |> Changeset.validate_required([:code, :description, :description_vector])
-  end
+  alias MedicalTranscription.Coding.CodeVector
 
   def insert_vector(params) do
-    changeset = changeset(%__MODULE__{}, params)
+    changeset =
+      %CodeVector{}
+      |> CodeVector.changeset(params)
 
     case Repo.insert(changeset) do
       {:ok, _} ->
@@ -42,20 +33,40 @@ defmodule MedicalTranscription.CodeVector do
 
   def exists_for_code?(code) do
     Repo.exists?(
-      from v in __MODULE__,
+      from v in CodeVector,
         where: v.code == ^code
     )
   end
 
   def icd9_present?() do
-    num_rows = Repo.aggregate(__MODULE__, :count)
+    num_rows = Repo.aggregate(CodeVector, :count)
 
     num_rows >= 14_567
   end
 
-  def find_similar(search_vector, limit \\ 5) do
+  def process_chunk(text, opts \\ []) do
+    k = Keyword.get(opts, :num_results, 5)
+
+    search_vector_for_db = compute_vector_as_list(text)
+
+    find_similar(search_vector_for_db, k)
+
+    # -- Remove matches that don't exceed a given threshold
+    # TODO: This depends on receiving a similarity score from the Postgres query
+    # |> Enum.filter(fn {_index, score} -> score >= similarity_threshold end)
+  end
+
+  @doc "Creates a vector embedding for text using the text embedding serving in the application's supervision tree."
+  def compute_vector_as_list(text) do
+    MedicalTranscription.TextEmbeddingServing
+    |> Nx.Serving.batched_run(text)
+    |> Map.get(:embedding)
+    |> Nx.to_flat_list()
+  end
+
+  defp find_similar(search_vector, limit \\ 5) do
     Repo.all(
-      from v in __MODULE__,
+      from v in CodeVector,
         order_by: cosine_distance(v.description_vector, ^search_vector),
         limit: ^limit
     )

lib/medical_transcription/coding/code_vector.ex

@@ -0,0 +1,19 @@
+defmodule MedicalTranscription.Coding.CodeVector do
+  @moduledoc """
+  Represents a code and its description, along with a vector embedding for its description.
+  """
+  use Ecto.Schema
+  alias Ecto.Changeset
+
+  schema "code_vectors" do
+    field :code, :string
+    field :description, :string
+    field :description_vector, Pgvector.Ecto.Vector
+  end
+
+  def changeset(code_vector, params \\ %{}) do
+    code_vector
+    |> Changeset.cast(params, [:code, :description, :description_vector])
+    |> Changeset.validate_required([:code, :description, :description_vector])
+  end
+end

lib/medical_transcription/coding/vector_precomputation.ex

@@ -1,7 +1,7 @@
 defmodule MedicalTranscription.Coding.VectorPrecomputation do
   @moduledoc "Populate database with vector embeddings from downloaded ICD-9 code list"
 
-  alias MedicalTranscription.CodeVector
+  alias MedicalTranscription.Coding
 
   @doc "Downloads the ICD-9 codelist, calculates vector embeddings for each, and adds them to the database"
   def run() do
@@ -31,7 +31,7 @@ defmodule MedicalTranscription.Coding.VectorPrecomputation do
     |> Stream.filter(fn %{"code" => code} -> String.length(code) > 0 end)
     |> Stream.with_index()
     |> Enum.each(fn {%{"code" => code, "long_description" => description}, index} ->
-      if !CodeVector.exists_for_code?(code) do
+      if !Coding.exists_for_code?(code) do
         compute_vector_for_code(code, description)
       end
 
@@ -60,9 +60,9 @@ defmodule MedicalTranscription.Coding.VectorPrecomputation do
   end
 
   defp compute_vector_for_code(code, description) do
-    vector_for_db = MedicalTranscription.Utilities.compute_vector_as_list(description)
+    vector_for_db = Coding.compute_vector_as_list(description)
 
-    CodeVector.insert_vector(%{
+    Coding.insert_vector(%{
       code: code,
       description: description,
       description_vector: vector_for_db

lib/medical_transcription/transcriber.ex

@@ -4,10 +4,10 @@ defmodule MedicalTranscription.Transcriber do
   to look for possible matching codes.
   """
 
-  alias MedicalTranscription.CodeSearcher
+  alias MedicalTranscription.Coding
 
   defp get_tags_and_send_result(chunk, index, live_view_pid) do
-    tags = CodeSearcher.process_chunk(chunk.text)
+    tags = Coding.process_chunk(chunk.text)
     result = build_result(index, chunk, tags)
 
     send(live_view_pid, {:transcription_row, result})

lib/medical_transcription/utilities.ex

@@ -1,13 +0,0 @@
-defmodule MedicalTranscription.Utilities do
-  @moduledoc """
-  Holds general utility functions. Over time, consider if there are other modules that make for a more relevant home.
-  """
-
-  @doc "Creates a vector embedding for text using the text embedding serving in the application's supervision tree."
-  def compute_vector_as_list(text) do
-    MedicalTranscription.TextEmbeddingServing
-    |> Nx.Serving.batched_run(text)
-    |> Map.get(:embedding)
-    |> Nx.to_flat_list()
-  end
-end

lib/medical_transcription_web/components/layouts/app.html.heex

@@ -14,7 +14,7 @@
     </div>
 
     <div class="px-6 flex flex-col items-center">
-      <%= if MedicalTranscription.CodeVector.icd9_present?() do %>
+      <%= if MedicalTranscription.Coding.icd9_present?() do %>
         <div
           class="w-full px-3 py-2 bg-emerald-600 text-white text-center rounded-lg"
           title="Precalculated vector embeddings for classification labels were found."

lib/medical_transcription_web/components/transcription_text_component.ex

@@ -6,7 +6,7 @@ defmodule MedicalTranscriptionWeb.Components.TranscriptionTextComponent do
   use MedicalTranscriptionWeb, :live_component
   import MedicalTranscriptionWeb.Components
   import MedicalTranscriptionWeb.Components.KeywordHighlighter
-  alias MedicalTranscription.CodeVector
+  alias MedicalTranscription.Coding.CodeVector
 
   @impl Phoenix.LiveComponent
   def update(assigns, socket) do

lib/medical_transcription_web/live/home_live/index.ex

@@ -82,7 +82,7 @@ defmodule MedicalTranscriptionWeb.HomeLive.Index do
 
   @impl true
   def handle_event("add_feedback", params, socket) do
-    text_vector = MedicalTranscription.Utilities.compute_vector_as_list(params["text"])
+    text_vector = MedicalTranscription.Coding.compute_vector_as_list(params["text"])
 
     result =
       params
@@ -135,7 +135,7 @@ defmodule MedicalTranscriptionWeb.HomeLive.Index do
 
   @impl true
   def handle_info({:received_audio_payload, transcribed_text}, socket) do
-    tags = MedicalTranscription.CodeSearcher.process_chunk(transcribed_text)
+    tags = MedicalTranscription.Coding.process_chunk(transcribed_text)
 
     result = %{
       id: socket.assigns.current_recording_id + 1,