refactor (v2): better arch

Co-authored-by: Copilot <copilot@github.com>

crates/adapters/fastembed/Cargo.toml

@@ -0,0 +1,14 @@
+[package]
+name = "fastembed-adapter"
+version = "0.1.0"
+edition = "2024"
+
+[dependencies]
+domain = { workspace = true }
+async-trait = { workspace = true }
+tokio = { workspace = true }
+tracing = { workspace = true }
+fastembed = "5"
+
+[dev-dependencies]
+tokio = { workspace = true }

crates/adapters/fastembed/src/lib.rs

@@ -0,0 +1,88 @@
+use std::{
+    path::PathBuf,
+    sync::{Arc, Mutex},
+};
+
+use async_trait::async_trait;
+use fastembed::{EmbeddingModel, TextEmbedding, TextInitOptions};
+
+use domain::{
+    errors::{DomainError, DomainResult},
+    smart::ports::EmbeddingGenerator,
+};
+
+pub struct FastEmbedConfig {
+    pub model: EmbeddingModel,
+    /// Directory used to cache downloaded model files.
+    /// Defaults to the system cache directory when `None`.
+    pub cache_dir: Option<PathBuf>,
+    pub show_download_progress: bool,
+}
+
+impl Default for FastEmbedConfig {
+    fn default() -> Self {
+        Self {
+            model: EmbeddingModel::AllMiniLML6V2,
+            cache_dir: None,
+            show_download_progress: false,
+        }
+    }
+}
+
+impl FastEmbedConfig {
+    pub fn with_model(model: EmbeddingModel) -> Self {
+        Self {
+            model,
+            ..Default::default()
+        }
+    }
+}
+
+pub struct FastEmbedGenerator {
+    model: Arc<Mutex<TextEmbedding>>,
+}
+
+impl FastEmbedGenerator {
+    /// Initialise the model. Downloads and caches model files on first call.
+    pub fn new(config: FastEmbedConfig) -> Result<Self, DomainError> {
+        let mut opts = TextInitOptions::new(config.model)
+            .with_show_download_progress(config.show_download_progress);
+
+        if let Some(dir) = config.cache_dir {
+            opts = opts.with_cache_dir(dir);
+        }
+
+        let model = TextEmbedding::try_new(opts)
+            .map_err(|e| DomainError::Infrastructure(format!("fastembed init failed: {e}")))?;
+
+        Ok(Self {
+            model: Arc::new(Mutex::new(model)),
+        })
+    }
+}
+
+#[async_trait]
+impl EmbeddingGenerator for FastEmbedGenerator {
+    async fn generate(&self, text: &str) -> DomainResult<Vec<f32>> {
+        let model = Arc::clone(&self.model);
+        let text = text.to_owned();
+
+        tokio::task::spawn_blocking(move || {
+            let mut guard = model
+                .lock()
+                .map_err(|_| DomainError::Infrastructure("model mutex poisoned".into()))?;
+            guard
+                .embed(vec![text.as_str()], None)
+                .map_err(|e| DomainError::Infrastructure(format!("embedding failed: {e}")))?
+                .into_iter()
+                .next()
+                .ok_or_else(|| DomainError::Infrastructure("no embedding returned".into()))
+        })
+        .await
+        .map_err(|e| DomainError::Infrastructure(format!("spawn_blocking panicked: {e}")))?
+    }
+}
+
+#[cfg(test)]
+#[path = "tests/lib.rs"]
+mod tests;

crates/adapters/fastembed/src/tests/lib.rs

@@ -0,0 +1,33 @@
+use crate::{FastEmbedConfig, FastEmbedGenerator};
+use domain::smart::ports::EmbeddingGenerator;
+use fastembed::EmbeddingModel;
+
+/// Downloads the model on first run (~90 MB). Run with:
+///   cargo test -p fastembed-adapter -- --ignored
+#[tokio::test]
+#[ignore]
+async fn generates_embedding_with_correct_dimension() {
+    let generator =
+        FastEmbedGenerator::new(FastEmbedConfig::with_model(EmbeddingModel::AllMiniLML6V2))
+            .expect("model init failed");
+
+    let embedding = generator.generate("hello world").await.unwrap();
+
+    // AllMiniLML6V2 produces 384-dimensional vectors.
+    assert_eq!(embedding.len(), 384);
+    // Sanity: values are in a reasonable range.
+    assert!(embedding.iter().all(|v| v.is_finite()));
+}
+
+#[tokio::test]
+#[ignore]
+async fn different_texts_produce_different_embeddings() {
+    let generator =
+        FastEmbedGenerator::new(FastEmbedConfig::with_model(EmbeddingModel::AllMiniLML6V2))
+            .expect("model init failed");
+
+    let a = generator.generate("cats").await.unwrap();
+    let b = generator.generate("quantum mechanics").await.unwrap();
+
+    assert_ne!(a, b);
+}