feat: scaffold FastAPI backend with ports-and-adapters layout

backend/app/domain/matching.py

@@ -0,0 +1,13 @@
+"""Fuzzy grounding: decide whether a search hit matches an LLM candidate."""
+
+from difflib import SequenceMatcher
+
+
+def normalize(text: str) -> str:
+    """Lowercase and strip decorations that vary across releases."""
+    return text.lower().strip()
+
+
+def title_similarity(candidate_title: str, catalog_title: str) -> float:
+    """Ratio in [0, 1] between a proposed title and a catalog title."""
+    return SequenceMatcher(None, normalize(candidate_title), normalize(catalog_title)).ratio()

backend/app/domain/models.py

@@ -0,0 +1,30 @@
+"""Core domain models — pure data, no app-level imports."""
+
+from pydantic import BaseModel
+
+
+class Track(BaseModel):
+    """A resolved, verified track from the catalog."""
+
+    spotify_id: str
+    title: str
+    artists: list[str]
+    album: str
+    album_art_url: str | None = None
+
+
+class TasteProfile(BaseModel):
+    """Compressed listening profile feeding both LLM calls."""
+
+    top_artists_long_term: list[str]
+    top_artists_short_term: list[str]
+    top_tracks_short_term: list[str]
+    saved_track_ids: set[str]
+
+
+class Recommendation(BaseModel):
+    """A ranked track with its one-line justification."""
+
+    track: Track
+    justification: str
+    rank: int

backend/app/domain/profile.py

@@ -0,0 +1,12 @@
+"""Taste-profile compression: raw listening data to a prompt-sized summary."""
+
+from app.domain.models import TasteProfile
+
+
+def compress_taste_profile(profile: TasteProfile, artist_limit: int = 15) -> str:
+    """Render a profile as compact prompt text, bounded in size."""
+    return (
+        f"Top artists (long term): {', '.join(profile.top_artists_long_term[:artist_limit])}. "
+        f"Top artists (recent): {', '.join(profile.top_artists_short_term[:artist_limit])}. "
+        f"Recent favourite tracks: {', '.join(profile.top_tracks_short_term[:artist_limit])}."
+    )