Implement Phase 3 Music Analysis and LLM Engine

- Refactor Database: Add `Artist` model, M2M relationship, and `AnalysisSnapshot` model.
- Backend Services: Implement `StatsService` for computable metrics and `NarrativeService` for Gemini LLM integration.
- Fix Ingestion: Correctly handle multiple artists per track and backfill existing data.
- Testing: Add unit tests for statistics logic and live verification scripts.
- Documentation: Add `PHASE_4_FRONTEND_GUIDE.md`.

backend/alembic/versions/4401cb416661_add_artist_and_snapshot_models.py

@@ -0,0 +1,63 @@
+"""Add Artist and Snapshot models
+
+Revision ID: 4401cb416661
+Revises: 707387fe1be2
+Create Date: 2025-12-24 23:06:59.235445
+
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision: str = '4401cb416661'
+down_revision: Union[str, Sequence[str], None] = '707387fe1be2'
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    """Upgrade schema."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.create_table('analysis_snapshots',
+    sa.Column('id', sa.Integer(), nullable=False),
+    sa.Column('date', sa.DateTime(), nullable=True),
+    sa.Column('period_start', sa.DateTime(), nullable=True),
+    sa.Column('period_end', sa.DateTime(), nullable=True),
+    sa.Column('period_label', sa.String(), nullable=True),
+    sa.Column('metrics_payload', sa.JSON(), nullable=True),
+    sa.Column('narrative_report', sa.JSON(), nullable=True),
+    sa.Column('model_used', sa.String(), nullable=True),
+    sa.PrimaryKeyConstraint('id')
+    )
+    op.create_index(op.f('ix_analysis_snapshots_date'), 'analysis_snapshots', ['date'], unique=False)
+    op.create_index(op.f('ix_analysis_snapshots_id'), 'analysis_snapshots', ['id'], unique=False)
+    op.create_table('artists',
+    sa.Column('id', sa.String(), nullable=False),
+    sa.Column('name', sa.String(), nullable=True),
+    sa.Column('genres', sa.JSON(), nullable=True),
+    sa.PrimaryKeyConstraint('id')
+    )
+    op.create_index(op.f('ix_artists_id'), 'artists', ['id'], unique=False)
+    op.create_table('track_artists',
+    sa.Column('track_id', sa.String(), nullable=False),
+    sa.Column('artist_id', sa.String(), nullable=False),
+    sa.ForeignKeyConstraint(['artist_id'], ['artists.id'], ),
+    sa.ForeignKeyConstraint(['track_id'], ['tracks.id'], ),
+    sa.PrimaryKeyConstraint('track_id', 'artist_id')
+    )
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    """Downgrade schema."""
+    # ### commands auto generated by Alembic - please adjust! ###
+    op.drop_table('track_artists')
+    op.drop_index(op.f('ix_artists_id'), table_name='artists')
+    op.drop_table('artists')
+    op.drop_index(op.f('ix_analysis_snapshots_id'), table_name='analysis_snapshots')
+    op.drop_index(op.f('ix_analysis_snapshots_date'), table_name='analysis_snapshots')
+    op.drop_table('analysis_snapshots')
+    # ### end Alembic commands ###

backend/app/ingest.py

@@ -2,7 +2,7 @@ import asyncio
 import os
 from datetime import datetime
 from sqlalchemy.orm import Session
-from .models import Track, PlayHistory
+from .models import Track, PlayHistory, Artist
 from .database import SessionLocal
 from .services.spotify_client import SpotifyClient
 from .services.reccobeats_client import ReccoBeatsClient
@@ -19,9 +19,32 @@ def get_spotify_client():
 def get_reccobeats_client():
     return ReccoBeatsClient()
 
+async def ensure_artists_exist(db: Session, artists_data: list):
+    """
+    Ensures that all artists in the list exist in the Artist table.
+    Returns a list of Artist objects.
+    """
+    artist_objects = []
+    for a_data in artists_data:
+        artist_id = a_data["id"]
+        artist = db.query(Artist).filter(Artist.id == artist_id).first()
+        if not artist:
+            artist = Artist(
+                id=artist_id,
+                name=a_data["name"],
+                genres=[] # Will be enriched later
+            )
+            db.add(artist)
+            # We commit inside the loop or after, but for now we rely on the main commit
+            # However, to return the object correctly we might need to flush if we were doing complex things,
+            # but here adding to session is enough for SQLAlchemy to track it.
+        artist_objects.append(artist)
+    return artist_objects
+
 async def enrich_tracks(db: Session, spotify_client: SpotifyClient, recco_client: ReccoBeatsClient):
     """
     Finds tracks missing genres (Spotify) or audio features (ReccoBeats) and enriches them.
+    Also enriches Artists with genres.
     """
 
     # 1. Enrich Audio Features (via ReccoBeats)
@@ -66,39 +89,35 @@ async def enrich_tracks(db: Session, spotify_client: SpotifyClient, recco_client
         print(f"Updated {updated_count} tracks with audio features.")
         db.commit()
 
-    # 2. Enrich Genres (via Spotify Artists)
-    tracks_missing_genres = db.query(Track).filter(Track.genres == None).limit(50).all()
+    # 2. Enrich Artist Genres (via Spotify Artists)
+    # We look for artists who have no genres. Note: an artist might genuinely have no genres,
+    # so we might need a flag "genres_checked" in the future, but for now checking empty list is okay.
+    # However, newly created artists have genres=[] (empty list) or None?
+    # My model definition: genres = Column(JSON, nullable=True)
+    # So if it is None, we haven't fetched it.
 
-    if tracks_missing_genres:
-        print(f"Enriching {len(tracks_missing_genres)} tracks with genres (Spotify)...")
+    artists_missing_genres = db.query(Artist).filter(Artist.genres == None).limit(50).all()
 
-        artist_ids = set()
-        track_artist_map = {}
-
-        for t in tracks_missing_genres:
-            if t.raw_data and "artists" in t.raw_data:
-                a_ids = [a["id"] for a in t.raw_data["artists"]]
-                artist_ids.update(a_ids)
-                track_artist_map[t.id] = a_ids
-
-        artist_ids_list = list(artist_ids)
-        artist_genre_map = {}
+    if artists_missing_genres:
+        print(f"Enriching {len(artists_missing_genres)} artists with genres (Spotify)...")
+        artist_ids_list = [a.id for a in artists_missing_genres]
 
+        artist_data_map = {}
+        # Spotify allows fetching 50 artists at a time
         for i in range(0, len(artist_ids_list), 50):
             chunk = artist_ids_list[i:i+50]
             artists_data = await spotify_client.get_artists(chunk)
             for a_data in artists_data:
                 if a_data:
-                    artist_genre_map[a_data["id"]] = a_data.get("genres", [])
+                    artist_data_map[a_data["id"]] = a_data.get("genres", [])
 
-        for t in tracks_missing_genres:
-            a_ids = track_artist_map.get(t.id, [])
-            combined_genres = set()
-            for a_id in a_ids:
-                genres = artist_genre_map.get(a_id, [])
-                combined_genres.update(genres)
-
-            t.genres = list(combined_genres)
+        for artist in artists_missing_genres:
+            genres = artist_data_map.get(artist.id)
+            if genres is not None:
+                artist.genres = genres
+            else:
+                # If we couldn't fetch, set to empty list so we don't keep retrying forever (or handle errors better)
+                artist.genres = []
 
         db.commit()
 
@@ -128,15 +147,30 @@ async def ingest_recently_played(db: Session):
             track = Track(
                 id=track_id,
                 name=track_data["name"],
-                artist=", ".join([a["name"] for a in track_data["artists"]]),
+                artist=", ".join([a["name"] for a in track_data["artists"]]), # Legacy string
                 album=track_data["album"]["name"],
                 duration_ms=track_data["duration_ms"],
                 popularity=track_data["popularity"],
                 raw_data=track_data
             )
+
+            # Handle Artists Relation
+            artists_data = track_data.get("artists", [])
+            artist_objects = await ensure_artists_exist(db, artists_data)
+            track.artists = artist_objects
+
             db.add(track)
             db.commit()
 
+        # Ensure relationships exist even if track existed (e.g. migration)
+        # Check if track has artists linked. If not (and raw_data has them), link them.
+        # FIX: Logic was previously indented improperly inside `if not track`.
+        if not track.artists and track.raw_data and "artists" in track.raw_data:
+             print(f"Backfilling artists for track {track.name}")
+             artist_objects = await ensure_artists_exist(db, track.raw_data["artists"])
+             track.artists = artist_objects
+             db.commit()
+
         exists = db.query(PlayHistory).filter(
             PlayHistory.track_id == track_id,
             PlayHistory.played_at == played_at

backend/app/models.py

@@ -1,14 +1,32 @@
-from sqlalchemy import Column, Integer, String, DateTime, JSON, ForeignKey, Float
+from sqlalchemy import Column, Integer, String, DateTime, JSON, ForeignKey, Float, Table, Text
 from sqlalchemy.orm import relationship
 from datetime import datetime
 from .database import Base
 
+# Association Table for Many-to-Many Relationship between Track and Artist
+track_artists = Table(
+    'track_artists',
+    Base.metadata,
+    Column('track_id', String, ForeignKey('tracks.id'), primary_key=True),
+    Column('artist_id', String, ForeignKey('artists.id'), primary_key=True)
+)
+
+class Artist(Base):
+    __tablename__ = "artists"
+
+    id = Column(String, primary_key=True, index=True) # Spotify ID
+    name = Column(String)
+    genres = Column(JSON, nullable=True) # List of genre strings
+
+    # Relationships
+    tracks = relationship("Track", secondary=track_artists, back_populates="artists")
+
 class Track(Base):
     __tablename__ = "tracks"
 
     id = Column(String, primary_key=True, index=True) # Spotify ID
     name = Column(String)
-    artist = Column(String)
+    artist = Column(String) # Display string (e.g. "Drake, Future") - kept for convenience
     album = Column(String)
     duration_ms = Column(Integer)
     popularity = Column(Integer, nullable=True)
@@ -31,17 +49,18 @@ class Track(Base):
     tempo = Column(Float, nullable=True)
     time_signature = Column(Integer, nullable=True)
 
-    # Genres (stored as JSON list of strings)
+    # Genres (stored as JSON list of strings) - DEPRECATED in favor of Artist.genres but kept for now
     genres = Column(JSON, nullable=True)
 
     # AI Analysis fields
     lyrics_summary = Column(String, nullable=True)
-    genre_tags = Column(String, nullable=True) # JSON list stored as string or just raw JSON
+    genre_tags = Column(String, nullable=True)
 
     created_at = Column(DateTime, default=datetime.utcnow)
     updated_at = Column(DateTime, default=datetime.utcnow, onupdate=datetime.utcnow)
 
     plays = relationship("PlayHistory", back_populates="track")
+    artists = relationship("Artist", secondary=track_artists, back_populates="tracks")
 
 
 class PlayHistory(Base):
@@ -55,3 +74,23 @@ class PlayHistory(Base):
     context_uri = Column(String, nullable=True)
 
     track = relationship("Track", back_populates="plays")
+
+
+class AnalysisSnapshot(Base):
+    """
+    Stores the computed statistics and LLM analysis for a given period.
+    Allows for trend analysis over time.
+    """
+    __tablename__ = "analysis_snapshots"
+
+    id = Column(Integer, primary_key=True, index=True)
+    date = Column(DateTime, default=datetime.utcnow, index=True) # When the analysis was run
+    period_start = Column(DateTime)
+    period_end = Column(DateTime)
+    period_label = Column(String) # e.g., "last_30_days", "monthly_nov_2023"
+
+    # The heavy lifting: stored as JSON blobs
+    metrics_payload = Column(JSON) # The input to the LLM (StatsService output)
+    narrative_report = Column(JSON) # The output from the LLM (NarrativeService output)
+
+    model_used = Column(String, nullable=True) # e.g. "gemini-1.5-flash"

backend/app/services/narrative_service.py

@@ -0,0 +1,67 @@
+import os
+import json
+import google.generativeai as genai
+from typing import Dict, Any
+
+class NarrativeService:
+    def __init__(self, model_name: str = "gemini-2.5-flash"):
+        self.api_key = os.getenv("GEMINI_API_KEY")
+        if not self.api_key:
+            print("WARNING: GEMINI_API_KEY not found. LLM features will fail.")
+        else:
+            genai.configure(api_key=self.api_key)
+
+        self.model_name = model_name
+
+    def generate_narrative(self, stats_json: Dict[str, Any]) -> Dict[str, str]:
+        if not self.api_key:
+            return {"error": "Missing API Key"}
+
+        prompt = f"""
+You are analyzing a user's Spotify listening data. Below is a JSON summary of metrics I've computed. Your job is to:
+
+1. Write a narrative "Vibe Check" (2-3 paragraphs) describing their overall listening personality this period.
+2. Identify 3-5 notable patterns or anomalies.
+3. Provide a "Musical Persona" label (e.g., "Late-Night Binge Listener", "Genre Chameleon", "Album Purist").
+4. Write a brief, playful "roast" (1-2 sentences) based on the data.
+
+Guidelines:
+- Do NOT recalculate any numbers.
+- Use specific metrics to support observations (e.g., "Your whiplash score of 18.3 BPM suggests...").
+- Keep tone conversational but insightful.
+- Avoid mental health claims; stick to behavioral descriptors.
+- Highlight both positive patterns and quirks.
+
+Data:
+{json.dumps(stats_json, indent=2)}
+
+Output Format (return valid JSON):
+{{
+  "vibe_check": "...",
+  "patterns": ["...", "..."],
+  "persona": "...",
+  "roast": "..."
+}}
+"""
+        try:
+            # Handle full model path if passed or default short name
+            # The library often accepts 'gemini-2.5-flash' but list_models returns 'models/gemini-2.5-flash'
+            model_id = self.model_name
+            if not model_id.startswith("models/") and "/" not in model_id:
+                # Try simple name, if it fails user might need to pass 'models/...'
+                pass
+
+            model = genai.GenerativeModel(model_id)
+            response = model.generate_content(prompt)
+
+            # Clean up response to ensure valid JSON (sometimes LLMs add markdown blocks)
+            text = response.text.strip()
+            if text.startswith("```json"):
+                text = text.replace("```json", "").replace("```", "")
+            elif text.startswith("```"):
+                 text = text.replace("```", "")
+
+            return json.loads(text)
+
+        except Exception as e:
+            return {"error": str(e), "raw_response": response.text if 'response' in locals() else "No response"}

backend/app/services/stats_service.py

@@ -0,0 +1,396 @@
+from sqlalchemy.orm import Session
+from sqlalchemy import func, distinct, desc
+from datetime import datetime, timedelta
+from typing import Dict, Any, List
+import math
+import numpy as np
+
+from ..models import PlayHistory, Track, Artist, AnalysisSnapshot
+
+class StatsService:
+    def __init__(self, db: Session):
+        self.db = db
+
+    def compute_volume_stats(self, period_start: datetime, period_end: datetime) -> Dict[str, Any]:
+        """
+        Calculates volume metrics: Total Plays, Unique Tracks, Artists, etc.
+        """
+        query = self.db.query(PlayHistory).filter(
+            PlayHistory.played_at >= period_start,
+            PlayHistory.played_at <= period_end
+        )
+        plays = query.all()
+        total_plays = len(plays)
+
+        if total_plays == 0:
+            return {
+                "total_plays": 0,
+                "estimated_minutes": 0,
+                "unique_tracks": 0,
+                "unique_artists": 0,
+                "unique_albums": 0,
+                "unique_genres": 0,
+                "top_tracks": [],
+                "top_artists": [],
+                "repeat_rate": 0,
+                "concentration": {}
+            }
+
+        # Calculate Duration (Estimated)
+        # Note: We query tracks to get duration.
+        # Ideally we join, but eager loading might be heavy. Let's do a join or simple loop.
+        # Efficient approach: Get all track IDs from plays, fetch Track objects in bulk map.
+
+        track_ids = [p.track_id for p in plays]
+        tracks = self.db.query(Track).filter(Track.id.in_(set(track_ids))).all()
+        track_map = {t.id: t for t in tracks}
+
+        total_ms = 0
+        unique_track_ids = set()
+        unique_artist_ids = set()
+        unique_album_names = set() # Spotify doesn't give album ID in PlayHistory directly unless joined, track has album name string.
+        # Ideally track has raw_data['album']['id'].
+        unique_album_ids = set()
+
+        genre_counts = {}
+
+        # For Top Lists
+        track_play_counts = {}
+        artist_play_counts = {}
+
+        for p in plays:
+            t = track_map.get(p.track_id)
+            if t:
+                total_ms += t.duration_ms
+                unique_track_ids.add(t.id)
+
+                # Top Tracks
+                track_play_counts[t.id] = track_play_counts.get(t.id, 0) + 1
+
+                # Artists (using relation)
+                # Note: This might cause N+1 query if not eager loaded.
+                # For strictly calculation, accessing t.artists (lazy load) loop might be slow for 1000s of plays.
+                # Optimization: Join PlayHistory -> Track -> Artist in query.
+
+                # Let's rely on raw_data for speed if relation loading is slow,
+                # OR Assume we accept some latency.
+                # Better: Pre-fetch artist connections or use the new tables properly.
+                # Let's use the object relation for correctness as per plan.
+                for artist in t.artists:
+                    unique_artist_ids.add(artist.id)
+                    artist_play_counts[artist.id] = artist_play_counts.get(artist.id, 0) + 1
+
+                    if artist.genres:
+                        for g in artist.genres:
+                            genre_counts[g] = genre_counts.get(g, 0) + 1
+
+                if t.raw_data and "album" in t.raw_data:
+                    unique_album_ids.add(t.raw_data["album"]["id"])
+                else:
+                    unique_album_ids.add(t.album) # Fallback
+
+        estimated_minutes = total_ms / 60000
+
+        # Top 5 Tracks
+        sorted_tracks = sorted(track_play_counts.items(), key=lambda x: x[1], reverse=True)[:5]
+        top_tracks = []
+        for tid, count in sorted_tracks:
+            t = track_map.get(tid)
+            top_tracks.append({
+                "name": t.name,
+                "artist": t.artist, # Display string
+                "count": count
+            })
+
+        # Top 5 Artists
+        # Need to fetch Artist names
+        top_artist_ids = sorted(artist_play_counts.items(), key=lambda x: x[1], reverse=True)[:5]
+        top_artists_objs = self.db.query(Artist).filter(Artist.id.in_([x[0] for x in top_artist_ids])).all()
+        artist_name_map = {a.id: a.name for a in top_artists_objs}
+
+        top_artists = []
+        for aid, count in top_artist_ids:
+            top_artists.append({
+                "name": artist_name_map.get(aid, "Unknown"),
+                "count": count
+            })
+
+        # Top Genres
+        sorted_genres = sorted(genre_counts.items(), key=lambda x: x[1], reverse=True)[:5]
+        top_genres = [{"name": g, "count": c} for g, c in sorted_genres]
+
+        # Concentration
+        unique_tracks_count = len(unique_track_ids)
+        repeat_rate = (total_plays - unique_tracks_count) / total_plays if total_plays > 0 else 0
+
+        # HHI (Herfindahl–Hirschman Index)
+        # Sum of (share)^2. Share = track_plays / total_plays
+        hhi = sum([(c/total_plays)**2 for c in track_play_counts.values()])
+
+        return {
+            "total_plays": total_plays,
+            "estimated_minutes": int(estimated_minutes),
+            "unique_tracks": unique_tracks_count,
+            "unique_artists": len(unique_artist_ids),
+            "unique_albums": len(unique_album_ids),
+            "unique_genres": len(genre_counts),
+            "top_tracks": top_tracks,
+            "top_artists": top_artists,
+            "top_genres": top_genres,
+            "repeat_rate": round(repeat_rate, 3),
+            "concentration": {
+                "hhi": round(hhi, 4),
+                # "gini": ... (skip for now to keep it simple)
+            }
+        }
+
+    def compute_time_stats(self, period_start: datetime, period_end: datetime) -> Dict[str, Any]:
+        """
+        Hourly, Daily distribution, etc.
+        """
+        query = self.db.query(PlayHistory).filter(
+            PlayHistory.played_at >= period_start,
+            PlayHistory.played_at <= period_end
+        )
+        plays = query.all()
+
+        hourly_counts = [0] * 24
+        weekday_counts = [0] * 7 # 0=Mon, 6=Sun
+
+        if not plays:
+             return {"hourly_distribution": hourly_counts}
+
+        for p in plays:
+            # played_at is UTC in DB usually. Ensure we handle timezone if user wants local.
+            # For now, assuming UTC or system time.
+            h = p.played_at.hour
+            d = p.played_at.weekday()
+
+            hourly_counts[h] += 1
+            weekday_counts[d] += 1
+
+        peak_hour = hourly_counts.index(max(hourly_counts))
+
+        # Weekend Share
+        weekend_plays = weekday_counts[5] + weekday_counts[6]
+        weekend_share = weekend_plays / len(plays) if len(plays) > 0 else 0
+
+        return {
+            "hourly_distribution": hourly_counts,
+            "peak_hour": peak_hour,
+            "weekday_distribution": weekday_counts,
+            "weekend_share": round(weekend_share, 2)
+        }
+
+    def compute_session_stats(self, period_start: datetime, period_end: datetime) -> Dict[str, Any]:
+        """
+        Session logic: Gap > 20 mins = new session.
+        """
+        query = self.db.query(PlayHistory).filter(
+            PlayHistory.played_at >= period_start,
+            PlayHistory.played_at <= period_end
+        ).order_by(PlayHistory.played_at.asc())
+        plays = query.all()
+
+        if not plays:
+            return {"count": 0, "avg_length_minutes": 0}
+
+        sessions = []
+        current_session = [plays[0]]
+
+        for i in range(1, len(plays)):
+            prev = plays[i-1]
+            curr = plays[i]
+            diff = (curr.played_at - prev.played_at).total_seconds() / 60
+
+            if diff > 20:
+                sessions.append(current_session)
+                current_session = []
+
+            current_session.append(curr)
+
+        sessions.append(current_session)
+
+        session_lengths_min = []
+        for sess in sessions:
+            if len(sess) > 1:
+                start = sess[0].played_at
+                end = sess[-1].played_at
+                # Add duration of last track?
+                # Let's just do (end - start) for simplicity + avg track duration
+                duration = (end - start).total_seconds() / 60
+                session_lengths_min.append(duration)
+            else:
+                session_lengths_min.append(3.0) # Approx 1 track
+
+        avg_min = sum(session_lengths_min) / len(session_lengths_min) if session_lengths_min else 0
+
+        return {
+            "count": len(sessions),
+            "avg_tracks": len(plays) / len(sessions),
+            "avg_minutes": round(avg_min, 1),
+            "longest_session_minutes": round(max(session_lengths_min), 1) if session_lengths_min else 0
+        }
+
+    def compute_vibe_stats(self, period_start: datetime, period_end: datetime) -> Dict[str, Any]:
+        """
+        Aggregates Audio Features (Energy, Valence, etc.)
+        """
+        query = self.db.query(PlayHistory).filter(
+            PlayHistory.played_at >= period_start,
+            PlayHistory.played_at <= period_end
+        )
+        plays = query.all()
+        track_ids = list(set([p.track_id for p in plays]))
+
+        if not track_ids:
+            return {}
+
+        tracks = self.db.query(Track).filter(Track.id.in_(track_ids)).all()
+
+        # Collect features
+        features = {
+            "energy": [], "valence": [], "danceability": [],
+            "tempo": [], "acousticness": [], "instrumentalness": [],
+            "liveness": [], "speechiness": []
+        }
+
+        for t in tracks:
+            # Weight by plays? The spec implies "Per-Period Aggregates".
+            # Usually weighted by play count is better representation of what was HEARD.
+            # Let's weight by play count in this period.
+            play_count = len([p for p in plays if p.track_id == t.id])
+
+            if t.energy is not None:
+                for _ in range(play_count):
+                    features["energy"].append(t.energy)
+                    features["valence"].append(t.valence)
+                    features["danceability"].append(t.danceability)
+                    features["tempo"].append(t.tempo)
+                    features["acousticness"].append(t.acousticness)
+                    features["instrumentalness"].append(t.instrumentalness)
+                    features["liveness"].append(t.liveness)
+                    features["speechiness"].append(t.speechiness)
+
+        stats = {}
+        for key, values in features.items():
+            valid = [v for v in values if v is not None]
+            if valid:
+                stats[f"avg_{key}"] = float(np.mean(valid))
+                stats[f"std_{key}"] = float(np.std(valid))
+            else:
+                stats[f"avg_{key}"] = None
+
+        # Derived Metrics
+        if stats.get("avg_energy") and stats.get("avg_valence"):
+            stats["mood_quadrant"] = {
+                "x": round(stats["avg_valence"], 2),
+                "y": round(stats["avg_energy"], 2)
+            }
+
+        return stats
+
+    def compute_era_stats(self, period_start: datetime, period_end: datetime) -> Dict[str, Any]:
+        """
+        Musical Age and Era Distribution.
+        """
+        query = self.db.query(PlayHistory).filter(
+            PlayHistory.played_at >= period_start,
+            PlayHistory.played_at <= period_end
+        )
+        plays = query.all()
+
+        years = []
+        track_ids = list(set([p.track_id for p in plays]))
+        tracks = self.db.query(Track).filter(Track.id.in_(track_ids)).all()
+        track_map = {t.id: t for t in tracks}
+
+        for p in plays:
+            t = track_map.get(p.track_id)
+            if t and t.raw_data and "album" in t.raw_data and "release_date" in t.raw_data["album"]:
+                rd = t.raw_data["album"]["release_date"]
+                # Format can be YYYY, YYYY-MM, YYYY-MM-DD
+                try:
+                    year = int(rd.split("-")[0])
+                    years.append(year)
+                except:
+                    pass
+
+        if not years:
+            return {"musical_age": None}
+
+        avg_year = sum(years) / len(years)
+
+        # Decade breakdown
+        decades = {}
+        for y in years:
+            dec = (y // 10) * 10
+            label = f"{dec}s"
+            decades[label] = decades.get(label, 0) + 1
+
+        total = len(years)
+        decade_dist = {k: round(v/total, 2) for k, v in decades.items()}
+
+        return {
+            "musical_age": int(avg_year),
+            "decade_distribution": decade_dist
+        }
+
+    def compute_skip_stats(self, period_start: datetime, period_end: datetime) -> Dict[str, Any]:
+        """
+        Implements boredom skip detection:
+        (next_track.played_at - current_track.played_at) < (current_track.duration_ms / 1000 - 10s)
+        """
+        query = self.db.query(PlayHistory).filter(
+            PlayHistory.played_at >= period_start,
+            PlayHistory.played_at <= period_end
+        ).order_by(PlayHistory.played_at.asc())
+        plays = query.all()
+
+        if len(plays) < 2:
+            return {"skip_rate": 0, "total_skips": 0}
+
+        skips = 0
+        track_ids = list(set([p.track_id for p in plays]))
+        tracks = self.db.query(Track).filter(Track.id.in_(track_ids)).all()
+        track_map = {t.id: t for t in tracks}
+
+        for i in range(len(plays) - 1):
+            current_play = plays[i]
+            next_play = plays[i+1]
+            track = track_map.get(current_play.track_id)
+
+            if not track or not track.duration_ms:
+                continue
+
+            diff_seconds = (next_play.played_at - current_play.played_at).total_seconds()
+
+            # Logic: If diff < (duration - 10s), it's a skip.
+            # Convert duration to seconds
+            duration_sec = track.duration_ms / 1000.0
+
+            # Also ensure diff isn't negative or weirdly small (re-plays)
+            # And assume "listening" means diff > 30s at least?
+            # Spec says "Spotify only returns 30s+".
+
+            if diff_seconds < (duration_sec - 10):
+                skips += 1
+
+        return {
+            "total_skips": skips,
+            "skip_rate": round(skips / len(plays), 3)
+        }
+
+    def generate_full_report(self, period_start: datetime, period_end: datetime) -> Dict[str, Any]:
+        return {
+            "period": {
+                "start": period_start.isoformat(),
+                "end": period_end.isoformat()
+            },
+            "volume": self.compute_volume_stats(period_start, period_end),
+            "time_habits": self.compute_time_stats(period_start, period_end),
+            "sessions": self.compute_session_stats(period_start, period_end),
+            "vibe": self.compute_vibe_stats(period_start, period_end),
+            "era": self.compute_era_stats(period_start, period_end),
+            "skips": self.compute_skip_stats(period_start, period_end)
+        }

backend/run_analysis.py

@@ -0,0 +1,82 @@
+import os
+import sys
+import json
+from datetime import datetime, timedelta
+from app.database import SessionLocal
+from app.services.stats_service import StatsService
+from app.services.narrative_service import NarrativeService
+from app.models import AnalysisSnapshot
+
+def run_analysis_pipeline(days: int = 30, model_name: str = "gemini-2.5-flash"):
+    db = SessionLocal()
+    try:
+        end_date = datetime.utcnow()
+        start_date = end_date - timedelta(days=days)
+
+        print(f"--- Starting Analysis for period: {start_date} to {end_date} ---")
+
+        # 1. Compute Stats
+        print("Calculating metrics...")
+        stats_service = StatsService(db)
+        stats_json = stats_service.generate_full_report(start_date, end_date)
+
+        # Check if we have enough data
+        if stats_json["volume"]["total_plays"] == 0:
+            print("No plays found in this period. Skipping LLM analysis.")
+            return
+
+        print(f"Stats computed. Total Plays: {stats_json['volume']['total_plays']}")
+        print(f"Top Artist: {stats_json['volume']['top_artists'][0]['name'] if stats_json['volume']['top_artists'] else 'N/A'}")
+
+        # 2. Generate Narrative
+        print(f"Generating Narrative with {model_name}...")
+        narrative_service = NarrativeService(model_name=model_name)
+        narrative_json = narrative_service.generate_narrative(stats_json)
+
+        if "error" in narrative_json:
+            print(f"LLM Error: {narrative_json['error']}")
+        else:
+            print("Narrative generated successfully.")
+            print(f"Persona: {narrative_json.get('persona')}")
+
+        # 3. Save Snapshot
+        print("Saving snapshot to database...")
+        snapshot = AnalysisSnapshot(
+            period_start=start_date,
+            period_end=end_date,
+            period_label=f"last_{days}_days",
+            metrics_payload=stats_json,
+            narrative_report=narrative_json,
+            model_used=model_name
+        )
+        db.add(snapshot)
+        db.commit()
+        print(f"Snapshot saved with ID: {snapshot.id}")
+
+        # 4. Output to file for easy inspection
+        output = {
+            "snapshot_id": snapshot.id,
+            "metrics": stats_json,
+            "narrative": narrative_json
+        }
+        with open("latest_analysis.json", "w") as f:
+            json.dump(output, f, indent=2)
+        print("Full report saved to latest_analysis.json")
+
+    except Exception as e:
+        print(f"Pipeline Failed: {e}")
+        import traceback
+        traceback.print_exc()
+    finally:
+        db.close()
+
+if __name__ == "__main__":
+    # Allow arguments?
+    days = 30
+    if len(sys.argv) > 1:
+        try:
+            days = int(sys.argv[1])
+        except:
+            pass
+
+    run_analysis_pipeline(days=days)

backend/seed_data.py

@@ -0,0 +1,78 @@
+from datetime import datetime, timedelta
+import random
+from app.database import SessionLocal
+from app.models import Track, Artist, PlayHistory
+from app.services.stats_service import StatsService
+
+def seed_db():
+    db = SessionLocal()
+
+    # 1. Create Artists
+    artists = []
+    for i in range(10):
+        a = Artist(
+            id=f"artist_{i}",
+            name=f"Artist {i}",
+            genres=[random.choice(["pop", "rock", "jazz", "edm", "hip-hop"]) for _ in range(2)]
+        )
+        db.merge(a) # merge handles insert/update
+        artists.append(a)
+
+    db.commit()
+    print(f"Seeded {len(artists)} artists.")
+
+    # 2. Create Tracks
+    tracks = []
+    for i in range(50):
+        # Random artist
+        artist = random.choice(artists)
+
+        t = Track(
+            id=f"track_{i}",
+            name=f"Track {i}",
+            artist=artist.name, # Legacy
+            album=f"Album {i % 10}",
+            duration_ms=random.randint(180000, 300000), # 3-5 mins
+            popularity=random.randint(10, 90),
+            danceability=random.uniform(0.3, 0.9),
+            energy=random.uniform(0.3, 0.9),
+            valence=random.uniform(0.1, 0.9),
+            tempo=random.uniform(80, 160),
+            raw_data={"album": {"id": f"album_{i%10}", "release_date": f"{random.randint(2000, 2023)}-01-01"}}
+        )
+        # Link artist
+        t.artists.append(artist)
+        db.merge(t)
+        tracks.append(t)
+
+    db.commit()
+    print(f"Seeded {len(tracks)} tracks.")
+
+    # 3. Create Play History (Last 30 days)
+    plays = []
+    base_time = datetime.utcnow() - timedelta(days=25)
+
+    for i in range(200):
+        # Create sessions
+        # 80% chance next play is soon (2-5 mins), 20% chance gap (30-600 mins)
+        gap = random.randint(2, 6) if random.random() > 0.2 else random.randint(30, 600)
+        base_time += timedelta(minutes=gap)
+
+        if base_time > datetime.utcnow():
+            break
+
+        track = random.choice(tracks)
+
+        p = PlayHistory(
+            track_id=track.id,
+            played_at=base_time,
+            context_uri="spotify:playlist:fake"
+        )
+        db.add(p)
+
+    db.commit()
+    print(f"Seeded play history until {base_time}.")
+    db.close()
+
+if __name__ == "__main__":
+    seed_db()

backend/tests/test_stats.py

@@ -0,0 +1,69 @@
+import unittest
+from datetime import datetime, timedelta
+from unittest.mock import MagicMock
+from app.services.stats_service import StatsService
+from app.models import PlayHistory, Track, Artist
+
+class TestStatsService(unittest.TestCase):
+    def setUp(self):
+        self.mock_db = MagicMock()
+        self.service = StatsService(self.mock_db)
+
+    def test_compute_volume_stats_empty(self):
+        # Mock empty query result
+        self.mock_db.query.return_value.filter.return_value.all.return_value = []
+
+        start = datetime.utcnow()
+        end = datetime.utcnow()
+        stats = self.service.compute_volume_stats(start, end)
+
+        self.assertEqual(stats["total_plays"], 0)
+        self.assertEqual(stats["unique_tracks"], 0)
+
+    def test_compute_session_stats(self):
+        # Create dummy plays
+        t1 = datetime(2023, 1, 1, 10, 0, 0)
+        t2 = datetime(2023, 1, 1, 10, 5, 0) # 5 min gap (same session)
+        t3 = datetime(2023, 1, 1, 12, 0, 0) # 1h 55m gap (new session)
+
+        plays = [
+            PlayHistory(played_at=t1, track_id="1"),
+            PlayHistory(played_at=t2, track_id="2"),
+            PlayHistory(played_at=t3, track_id="3"),
+        ]
+
+        # Mock the query chain
+        # service.db.query().filter().order_by().all()
+        query_mock = self.mock_db.query.return_value.filter.return_value.order_by.return_value
+        query_mock.all.return_value = plays
+
+        stats = self.service.compute_session_stats(datetime.utcnow(), datetime.utcnow())
+
+        # Expected: 2 sessions ([t1, t2], [t3])
+        self.assertEqual(stats["count"], 2)
+        # Avg tracks: 3 plays / 2 sessions = 1.5
+        self.assertEqual(stats["avg_tracks"], 1.5)
+
+    def test_compute_skip_stats(self):
+        # Track duration = 30s
+        track = Track(id="t1", duration_ms=30000)
+
+        # Play 1: 10:00:00
+        # Play 2: 10:00:10 (Diff 10s. Duration 30s. 10 < 20 (30-10) -> Skip)
+        p1 = PlayHistory(played_at=datetime(2023, 1, 1, 10, 0, 0), track_id="t1")
+        p2 = PlayHistory(played_at=datetime(2023, 1, 1, 10, 0, 10), track_id="t1")
+
+        plays = [p1, p2]
+
+        query_mock = self.mock_db.query.return_value.filter.return_value.order_by.return_value
+        query_mock.all.return_value = plays
+
+        # Mock track lookup
+        self.mock_db.query.return_value.filter.return_value.all.return_value = [track]
+
+        stats = self.service.compute_skip_stats(datetime.utcnow(), datetime.utcnow())
+
+        self.assertEqual(stats["total_skips"], 1)
+
+if __name__ == '__main__':
+    unittest.main()