diff --git a/plots/spectrogram-mel/implementations/plotnine.py b/plots/spectrogram-mel/implementations/plotnine.py
new file mode 100644
index 0000000000..76126b4801
--- /dev/null
+++ b/plots/spectrogram-mel/implementations/plotnine.py
@@ -0,0 +1,189 @@
+""" pyplots.ai
+spectrogram-mel: Mel-Spectrogram for Audio Analysis
+Library: plotnine 0.15.3 | Python 3.14.3
+Quality: 90/100 | Created: 2026-03-11
+"""
+
+import numpy as np
+import pandas as pd
+from plotnine import (
+    aes,
+    coord_cartesian,
+    element_blank,
+    element_rect,
+    element_text,
+    geom_raster,
+    geom_segment,
+    geom_text,
+    ggplot,
+    guide_colorbar,
+    guides,
+    labs,
+    scale_fill_gradientn,
+    scale_x_continuous,
+    scale_y_continuous,
+    theme,
+    theme_minimal,
+)
+from scipy.signal import stft
+
+
+# Data - synthesize a 3-second audio signal with speech-like frequency components
+np.random.seed(42)
+sample_rate = 22050
+duration = 3.0
+n_samples = int(sample_rate * duration)
+t = np.linspace(0, duration, n_samples, endpoint=False)
+
+# Build a rich audio signal: fundamental + harmonics with time-varying amplitude
+fundamental = 220
+signal = (
+    0.6 * np.sin(2 * np.pi * fundamental * t) * np.exp(-0.3 * t)
+    + 0.4 * np.sin(2 * np.pi * 440 * t) * (0.5 + 0.5 * np.sin(2 * np.pi * 1.5 * t))
+    + 0.3 * np.sin(2 * np.pi * 880 * t) * np.exp(-0.5 * t)
+    + 0.2 * np.sin(2 * np.pi * 1320 * t) * (1 - t / duration)
+    + 0.15 * np.sin(2 * np.pi * 3300 * t) * np.exp(-1.0 * t)
+    + 0.1 * np.random.randn(n_samples) * np.exp(-0.8 * t)
+)
+
+# Add a frequency sweep (chirp) from 500 to 4000 Hz in the middle section
+chirp_mask = (t > 0.8) & (t < 2.0)
+chirp_freq = 500 + (4000 - 500) * (t[chirp_mask] - 0.8) / 1.2
+signal[chirp_mask] += 0.35 * np.sin(2 * np.pi * np.cumsum(chirp_freq) / sample_rate)
+
+# STFT
+n_fft = 2048
+hop_length = 512
+_, time_bins, Zxx = stft(signal, fs=sample_rate, nperseg=n_fft, noverlap=n_fft - hop_length)
+power_spec = np.abs(Zxx) ** 2
+
+# Mel filterbank
+n_mels = 128
+freq_bins = np.linspace(0, sample_rate / 2, power_spec.shape[0])
+
+mel_low = 2595.0 * np.log10(1.0 + 0 / 700.0)
+mel_high = 2595.0 * np.log10(1.0 + (sample_rate / 2) / 700.0)
+mel_points = np.linspace(mel_low, mel_high, n_mels + 2)
+hz_points = 700.0 * (10.0 ** (mel_points / 2595.0) - 1.0)
+
+# Vectorized mel filterbank using numpy broadcasting
+lower = hz_points[:-2, np.newaxis]  # (n_mels, 1)
+center = hz_points[1:-1, np.newaxis]  # (n_mels, 1)
+upper = hz_points[2:, np.newaxis]  # (n_mels, 1)
+freqs = freq_bins[np.newaxis, :]  # (1, n_freq)
+
+rising = np.where((freqs >= lower) & (freqs <= center) & (center != lower), (freqs - lower) / (center - lower), 0.0)
+falling = np.where((freqs > center) & (freqs <= upper) & (upper != center), (upper - freqs) / (upper - center), 0.0)
+filterbank = rising + falling
+
+# Apply mel filterbank and convert to dB
+mel_spec = filterbank @ power_spec
+mel_spec_db = 10 * np.log10(np.maximum(mel_spec, 1e-10))
+mel_spec_db -= mel_spec_db.max()
+
+# Build long-form DataFrame with evenly-spaced mel band indices for smooth raster
+mel_center_freqs = 700.0 * (10.0 ** (mel_points[1:-1] / 2595.0) - 1.0)
+time_grid, mel_idx_grid = np.meshgrid(time_bins, np.arange(n_mels))
+
+df = pd.DataFrame({"Time (s)": time_grid.ravel(), "mel_band": mel_idx_grid.ravel(), "Power (dB)": mel_spec_db.ravel()})
+
+# Y-axis tick positions: map Hz values to mel band indices
+y_ticks_hz = [128, 256, 512, 1024, 2048, 4096, 8000]
+y_ticks_hz = [f for f in y_ticks_hz if f <= sample_rate / 2]
+# Convert Hz to mel band index via interpolation
+y_ticks_band = np.interp(y_ticks_hz, mel_center_freqs, np.arange(n_mels))
+
+
+# Annotation data — grammar-of-graphics approach: data-driven geom layers
+f0_band = float(np.interp(220, mel_center_freqs, np.arange(n_mels)))
+h3_band = float(np.interp(880, mel_center_freqs, np.arange(n_mels)))
+
+df_labels = pd.DataFrame(
+    {"x": [2.85, 2.85], "y": [f0_band, h3_band], "label": ["F\u2080", "3rd"], "color": ["#fcffa4", "#fb9b06"]}
+)
+df_reflines = pd.DataFrame(
+    {"x": [0.0, 0.0], "xend": [duration, duration], "y": [f0_band, h3_band], "yend": [f0_band, h3_band]}
+)
+
+# Plot — geom_raster for smooth spectrogram, data-driven geom_text/geom_segment for annotations
+plot = (
+    ggplot(df, aes(x="Time (s)", y="mel_band", fill="Power (dB)"))
+    + geom_raster(interpolate=True)
+    + scale_fill_gradientn(
+        colors=[
+            "#000004",
+            "#1b0c41",
+            "#4a0c6b",
+            "#781c6d",
+            "#a52c60",
+            "#cf4446",
+            "#ed6925",
+            "#fb9b06",
+            "#f7d13d",
+            "#fcffa4",
+        ],
+        name="Power (dB)",
+    )
+    + guides(fill=guide_colorbar(nbin=256, display="raster"))
+    + geom_text(
+        aes(x="x", y="y", label="label"),
+        data=df_labels.iloc[[0]],
+        inherit_aes=False,
+        color="#fcffa4",
+        size=11,
+        ha="right",
+        fontweight="bold",
+        alpha=0.85,
+    )
+    + geom_text(
+        aes(x="x", y="y", label="label"),
+        data=df_labels.iloc[[1]],
+        inherit_aes=False,
+        color="#fb9b06",
+        size=9,
+        ha="right",
+        alpha=0.7,
+    )
+    + geom_segment(
+        aes(x="x", xend="xend", y="y", yend="yend"),
+        data=df_reflines.iloc[[0]],
+        inherit_aes=False,
+        color="#fcffa4",
+        alpha=0.15,
+        size=0.4,
+    )
+    + geom_segment(
+        aes(x="x", xend="xend", y="y", yend="yend"),
+        data=df_reflines.iloc[[1]],
+        inherit_aes=False,
+        color="#fb9b06",
+        alpha=0.12,
+        size=0.3,
+    )
+    + scale_x_continuous(expand=(0, 0))
+    + scale_y_continuous(breaks=y_ticks_band.tolist(), labels=[str(f) for f in y_ticks_hz], expand=(0, 0))
+    + coord_cartesian(ylim=(0, n_mels - 1))
+    + labs(x="Time (s)", y="Frequency (Hz)", title="spectrogram-mel \u00b7 plotnine \u00b7 pyplots.ai")
+    + theme_minimal()
+    + theme(
+        figure_size=(16, 9),
+        text=element_text(family="sans-serif"),
+        plot_title=element_text(size=24, ha="center", weight="bold", color="#e0e0e0", margin={"b": 8}),
+        axis_title_x=element_text(size=20, color="#cccccc", margin={"t": 10}),
+        axis_title_y=element_text(size=20, color="#cccccc", margin={"r": 8}),
+        axis_text_x=element_text(size=16, color="#aaaaaa"),
+        axis_text_y=element_text(size=16, color="#aaaaaa"),
+        legend_title=element_text(size=16, weight="bold", color="#cccccc"),
+        legend_text=element_text(size=14, color="#aaaaaa"),
+        legend_position="right",
+        legend_key_height=60,
+        legend_key_width=14,
+        panel_grid_major=element_blank(),
+        panel_grid_minor=element_blank(),
+        panel_background=element_rect(fill="#000004", color="none"),
+        plot_background=element_rect(fill="#0e0e1a", color="none"),
+        plot_margin=0.02,
+    )
+)
+
+plot.save("plot.png", dpi=300, verbose=False)
diff --git a/plots/spectrogram-mel/metadata/plotnine.yaml b/plots/spectrogram-mel/metadata/plotnine.yaml
new file mode 100644
index 0000000000..f991c2cdc2
--- /dev/null
+++ b/plots/spectrogram-mel/metadata/plotnine.yaml
@@ -0,0 +1,236 @@
+library: plotnine
+specification_id: spectrogram-mel
+created: '2026-03-11T19:40:09Z'
+updated: '2026-03-11T20:10:00Z'
+generated_by: claude-opus-4-5-20251101
+workflow_run: 22970857471
+issue: 4672
+python_version: 3.14.3
+library_version: 0.15.3
+preview_url: https://storage.googleapis.com/pyplots-images/plots/spectrogram-mel/plotnine/plot.png
+preview_thumb: https://storage.googleapis.com/pyplots-images/plots/spectrogram-mel/plotnine/plot_thumb.png
+preview_html: null
+quality_score: 90
+review:
+  strengths:
+  - Cohesive dark theme design with panel background matching colormap floor and harmonized
+    outer background
+  - Rich synthesized audio data with harmonics, chirp sweep, and amplitude decay creating
+    a visually compelling spectrogram
+  - Clean vectorized mel filterbank implementation avoiding librosa dependency
+  - Data-driven annotations using separate DataFrames following grammar-of-graphics
+    principles
+  - Full spec compliance with all required features correctly implemented
+  weaknesses:
+  - Library Mastery remains the main gap — implementation uses plotnine correctly
+    but does not leverage truly distinctive features like stat transformations or
+    custom geoms
+  - Annotation reference lines are extremely subtle (alpha 0.12-0.15), nearly invisible
+    in the final render
+  image_description: The plot displays a mel-spectrogram on a fully dark-themed canvas.
+    The x-axis shows time (0-3 seconds) and the y-axis shows mel-scaled frequency
+    with Hz labels at key mel band edges (128, 256, 512, 1024, 2048, 4096, 8000 Hz).
+    The colormap is a custom inferno-like gradient progressing from very dark black/purple
+    (low power ~-60 dB) through magenta and orange to bright yellow (0 dB). Clear
+    horizontal bands of energy are visible at harmonic frequencies (~220, 440, 880
+    Hz), with lower harmonics being brightest and decaying over time. A prominent
+    chirp sweep arcs from approximately 500 Hz at t=0.8s up to ~4000 Hz at t=2.0s,
+    creating a striking curved bright line that serves as the main visual focal point.
+    Two data-driven annotations label the fundamental frequency (F0 in bright yellow)
+    and 3rd harmonic (in orange), with very subtle reference lines at those frequencies.
+    A tall colorbar on the right is labeled Power (dB). The title reads spectrogram-mel
+    plotnine pyplots.ai in bold light gray. The panel background (#000004) matches
+    the colormap floor, and the outer plot background (#0e0e1a) creates a cohesive
+    dark theme throughout.
+  criteria_checklist:
+    visual_quality:
+      score: 29
+      max: 30
+      items:
+      - id: VQ-01
+        name: Text Legibility
+        score: 8
+        max: 8
+        passed: true
+        comment: 'All font sizes explicitly set: title=24pt, axis titles=20pt, tick
+          labels=16pt, legend title=16pt, legend text=14pt'
+      - id: VQ-02
+        name: No Overlap
+        score: 6
+        max: 6
+        passed: true
+        comment: All text fully readable, y-axis labels well-spaced, annotations avoid
+          collision
+      - id: VQ-03
+        name: Element Visibility
+        score: 6
+        max: 6
+        passed: true
+        comment: geom_raster with interpolate=True produces smooth gap-free spectrogram
+      - id: VQ-04
+        name: Color Accessibility
+        score: 4
+        max: 4
+        passed: true
+        comment: Custom inferno-like sequential colormap is perceptually uniform and
+          colorblind-safe
+      - id: VQ-05
+        name: Layout & Canvas
+        score: 3
+        max: 4
+        passed: true
+        comment: Good 16:9 proportions with cohesive dark background, slight right-side
+          weight from colorbar
+      - id: VQ-06
+        name: Axis Labels & Title
+        score: 2
+        max: 2
+        passed: true
+        comment: Time (s) and Frequency (Hz) with units, correct title format
+    design_excellence:
+      score: 15
+      max: 20
+      items:
+      - id: DE-01
+        name: Aesthetic Sophistication
+        score: 6
+        max: 8
+        passed: true
+        comment: Custom 10-color gradient, dark panel matching colormap floor, cohesive
+          dark outer background, text color hierarchy
+      - id: DE-02
+        name: Visual Refinement
+        score: 5
+        max: 6
+        passed: true
+        comment: Grid removed, dark theme throughout, generous legend key, custom
+          margins. Reference lines nearly invisible
+      - id: DE-03
+        name: Data Storytelling
+        score: 4
+        max: 6
+        passed: true
+        comment: F0 and 3rd harmonic annotations guide viewer, chirp sweep creates
+          focal point, time-varying decay adds narrative
+    spec_compliance:
+      score: 15
+      max: 15
+      items:
+      - id: SC-01
+        name: Plot Type
+        score: 5
+        max: 5
+        passed: true
+        comment: Correct mel-spectrogram with STFT, mel filterbank, and dB conversion
+      - id: SC-02
+        name: Required Features
+        score: 4
+        max: 4
+        passed: true
+        comment: 'All spec features present: dB scale, sequential colormap, mel y-axis
+          with Hz labels, colorbar, correct parameters'
+      - id: SC-03
+        name: Data Mapping
+        score: 3
+        max: 3
+        passed: true
+        comment: X=time, Y=mel frequency with Hz labels, fill=power in dB
+      - id: SC-04
+        name: Title & Legend
+        score: 3
+        max: 3
+        passed: true
+        comment: Correct title format, colorbar labeled Power (dB)
+    data_quality:
+      score: 15
+      max: 15
+      items:
+      - id: DQ-01
+        name: Feature Coverage
+        score: 6
+        max: 6
+        passed: true
+        comment: Rich signal with fundamental, harmonics, chirp sweep, noise, time-varying
+          amplitudes
+      - id: DQ-02
+        name: Realistic Context
+        score: 5
+        max: 5
+        passed: true
+        comment: 220 Hz fundamental with natural harmonics is realistic for speech/music
+          analysis
+      - id: DQ-03
+        name: Appropriate Scale
+        score: 4
+        max: 4
+        passed: true
+        comment: Standard 22050 Hz sample rate, 3-second duration, standard STFT parameters
+    code_quality:
+      score: 10
+      max: 10
+      items:
+      - id: CQ-01
+        name: KISS Structure
+        score: 3
+        max: 3
+        passed: true
+        comment: Clean linear flow with no functions or classes
+      - id: CQ-02
+        name: Reproducibility
+        score: 2
+        max: 2
+        passed: true
+        comment: np.random.seed(42) set
+      - id: CQ-03
+        name: Clean Imports
+        score: 2
+        max: 2
+        passed: true
+        comment: All imports used, scipy.signal.stft appropriate
+      - id: CQ-04
+        name: Code Elegance
+        score: 2
+        max: 2
+        passed: true
+        comment: Vectorized mel filterbank, data-driven annotations are elegant
+      - id: CQ-05
+        name: Output & API
+        score: 1
+        max: 1
+        passed: true
+        comment: Saves as plot.png with dpi=300, current API
+    library_mastery:
+      score: 6
+      max: 10
+      items:
+      - id: LM-01
+        name: Idiomatic Usage
+        score: 4
+        max: 5
+        passed: true
+        comment: Strong grammar-of-graphics composition with data-driven annotations
+          via separate DataFrames
+      - id: LM-02
+        name: Distinctive Features
+        score: 2
+        max: 5
+        passed: false
+        comment: geom_raster(interpolate=True) and guide_colorbar are plotnine-specific
+          but not truly distinctive
+  verdict: APPROVED
+impl_tags:
+  dependencies:
+  - scipy
+  techniques:
+  - colorbar
+  - annotations
+  - layer-composition
+  patterns:
+  - data-generation
+  - matrix-construction
+  dataprep:
+  - binning
+  styling:
+  - dark-theme
+  - custom-colormap
+  - alpha-blending