Create container image and move parakeet fork to vendor dir

2 files changed, 46 insertions, 9 deletions

diff --git a/makima/src/bin/server.rs b/makima/src/bin/server.rs
index 06b6585..470e295 100644
--- a/makima/src/bin/server.rs
+++ b/makima/src/bin/server.rs
@@ -8,10 +8,10 @@ use tracing_subscriber::{layer::SubscriberExt, util::SubscriberInitExt};
 
 use makima::server::{run_server, state::AppState};
 
-/// Default model paths relative to the working directory.
-const PARAKEET_MODEL_DIR: &str = "models/parakeet-tdt-0.6b-v3";
-const PARAKEET_EOU_DIR: &str = "models/realtime_eou_120m-v1-onnx";
-const SORTFORMER_MODEL_PATH: &str = "models/diarization/diar_streaming_sortformer_4spk-v2.onnx";
+/// Default model paths (can be overridden via environment variables).
+const DEFAULT_PARAKEET_MODEL_DIR: &str = "models/parakeet-tdt-0.6b-v3";
+const DEFAULT_PARAKEET_EOU_DIR: &str = "models/realtime_eou_120m-v1-onnx";
+const DEFAULT_SORTFORMER_MODEL_PATH: &str = "models/diarization/diar_streaming_sortformer_4spk-v2.onnx";
 
 #[tokio::main]
 async fn main() -> anyhow::Result<()> {
@@ -19,22 +19,38 @@ async fn main() -> anyhow::Result<()> {
     tracing_subscriber::registry()
         .with(
             tracing_subscriber::EnvFilter::try_from_default_env()
-                .unwrap_or_else(|_| "makima=debug,tower_http=debug".into()),
+                .unwrap_or_else(|_| "makima=info,tower_http=info".into()),
         )
         .with(tracing_subscriber::fmt::layer())
         .init();
 
     tracing::info!("Starting Makima Listening API Server");
-    tracing::info!("Loading ML models...");
+
+    // Read configuration from environment
+    let port = std::env::var("PORT").unwrap_or_else(|_| "8080".to_string());
+    let parakeet_dir = std::env::var("PARAKEET_MODEL_DIR")
+        .unwrap_or_else(|_| DEFAULT_PARAKEET_MODEL_DIR.to_string());
+    let parakeet_eou_dir = std::env::var("PARAKEET_EOU_DIR")
+        .unwrap_or_else(|_| DEFAULT_PARAKEET_EOU_DIR.to_string());
+    let sortformer_path = std::env::var("SORTFORMER_MODEL_PATH")
+        .unwrap_or_else(|_| DEFAULT_SORTFORMER_MODEL_PATH.to_string());
+
+    tracing::info!(
+        parakeet = %parakeet_dir,
+        eou = %parakeet_eou_dir,
+        sortformer = %sortformer_path,
+        "Loading ML models..."
+    );
 
     // Load ML models
     let state = Arc::new(
-        AppState::new(PARAKEET_MODEL_DIR, PARAKEET_EOU_DIR, SORTFORMER_MODEL_PATH)
+        AppState::new(&parakeet_dir, &parakeet_eou_dir, &sortformer_path)
             .map_err(|e| anyhow::anyhow!("Failed to load models: {}", e))?,
     );
 
     tracing::info!("Models loaded successfully");
 
     // Run the server
-    run_server(state, "0.0.0.0:8080").await
+    let addr = format!("0.0.0.0:{}", port);
+    run_server(state, &addr).await
 }
diff --git a/makima/src/server/mod.rs b/makima/src/server/mod.rs
index a6e0525..c509afa 100644
--- a/makima/src/server/mod.rs
+++ b/makima/src/server/mod.rs
@@ -6,9 +6,12 @@ pub mod openapi;
 pub mod state;
 
 use axum::{
+    http::StatusCode,
+    response::IntoResponse,
     routing::get,
-    Router,
+    Json, Router,
 };
+use serde::Serialize;
 use tower_http::cors::{Any, CorsLayer};
 use tower_http::trace::TraceLayer;
 use utoipa::OpenApi;
@@ -18,6 +21,23 @@ use crate::server::handlers::listen;
 use crate::server::openapi::ApiDoc;
 use crate::server::state::SharedState;
 
+#[derive(Serialize)]
+struct HealthResponse {
+    status: &'static str,
+    version: &'static str,
+}
+
+/// Health check endpoint for load balancers and orchestrators.
+async fn health_check() -> impl IntoResponse {
+    (
+        StatusCode::OK,
+        Json(HealthResponse {
+            status: "healthy",
+            version: env!("CARGO_PKG_VERSION"),
+        }),
+    )
+}
+
 /// Create the axum Router with all routes configured.
 pub fn make_router(state: SharedState) -> Router {
     // API v1 routes
@@ -29,6 +49,7 @@ pub fn make_router(state: SharedState) -> Router {
         .url("/api-docs/openapi.json", ApiDoc::openapi());
 
     Router::new()
+        .route("/api/v1/healthcheck", get(health_check))
         .nest("/api/v1", api_v1)
         .merge(swagger)
         .layer(