diff --git a/notebooks_examples/batch/__init__.py b/notebooks_examples/batch/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/notebooks_examples/batch/batch-music.py b/notebooks_examples/batch/batch-music.py
deleted file mode 100644
index 5eb9e47..0000000
--- a/notebooks_examples/batch/batch-music.py
+++ /dev/null
@@ -1,84 +0,0 @@
-# ---
-# jupyter:
-#   jupytext:
-#     cell_metadata_filter: -all
-#     custom_cell_magics: kql
-#     text_representation:
-#       extension: .py
-#       format_name: percent
-#       format_version: '1.3'
-#       jupytext_version: 1.11.2
-#   kernelspec:
-#     display_name: process-improve
-#     language: python
-#     name: python3
-# ---
-
-# %%
-import os
-import pathlib
-import sys
-
-import numpy as np
-import pandas as pd
-
-# Some basic checks, and logging set up.
-while not str(pathlib.Path.cwd().name).lower().startswith("process-improve"):
-    os.chdir(pathlib.Path.cwd().parents[0])
-basecwd = pathlib.Path.cwd()
-sys.path.insert(0, str(basecwd))
-assert basecwd.exists()
-
-import process_improve.datasets.batch as batch_ds
-from process_improve.batch.data_input import melted_to_dict
-from process_improve.batch.plotting import plot_all_batches_per_tag
-
-pd.options.plotting.backend = "plotly"
-pd.options.display.max_columns = 20
-pd.options.display.width = 200
-
-
-# %%
-# Import the data: a dictionary of dataframes
-
-dryer_raw = pd.read_csv(pathlib.Path(batch_ds.__path__._recalculate()[0]) / "dryer.csv")
-df_dict = melted_to_dict(dryer_raw, batch_id_col="batch_id")
-
-# %%
-n_levels = 180
-percentile_low, percentile_high = 0.05, 0.95  # we create a range of n_levels between this low and high value
-
-# Ideally, use more than 1 tag to align on. These columns must exist in all data frames for all batches.
-columns_to_import = ["AgitatorPower", "AgitatorTorque", "JacketTemperature", "DryerTemp"]
-tag_to_plot = columns_to_import[2]
-
-# %%
-# Plot some data, to get an idea of what is present
-plot_all_batches_per_tag(df_dict=df_dict, tag=tag_to_plot, time_column="ClockTime", x_axis_label="Time [hours]")
-
-
-# %%
-for column in columns_to_import:
-    series = dryer_raw[column]
-    percentiles = series.quantile([percentile_low, percentile_high])
-
-    # Create a range of n_levels between the low and high percentile. then map the data to these levels,
-    # but don't clip the values; rather allow them to be outside the range, even if negative.
-
-    bins, step_size = np.linspace(percentiles[percentile_low], percentiles[percentile_high], n_levels + 1, retstep=True)
-    # Extend the bins to the left and right to allow for values outside the range, in the same step size
-    all_bins = np.concatenate(
-        [
-            np.flip(np.arange(start=bins[0], stop=series.min() - step_size, step=-step_size))[:-1],
-            bins,
-            np.arange(start=bins[-1], stop=series.max() + step_size, step=+step_size)[1:],
-        ]
-    )
-    levels = pd.cut(series, bins=all_bins, labels=range(len(all_bins) - 1), include_lowest=True)
-
-
-# %%
-# Next steps: graph storage of the nodes, and then create a graph of the transitions between the nodes.
-# Handle pulses.
-
-# %%
diff --git a/notebooks_examples/batch/batch_music.py b/notebooks_examples/batch/batch_music.py
new file mode 100644
index 0000000..9976c47
--- /dev/null
+++ b/notebooks_examples/batch/batch_music.py
@@ -0,0 +1,186 @@
+# ---
+# jupyter:
+#   jupytext:
+#     cell_metadata_filter: -all
+#     custom_cell_magics: kql
+#     text_representation:
+#       extension: .py
+#       format_name: percent
+#       format_version: '1.3'
+#       jupytext_version: 1.11.2
+#   kernelspec:
+#     display_name: process-improve
+#     language: python
+#     name: python3
+# ---
+
+# %%
+from __future__ import annotations
+
+import os
+import pathlib
+import sys
+
+import numpy as np
+import pandas as pd
+from neomodel import (
+    DateTimeProperty,
+    FloatProperty,
+    IntegerProperty,
+    RelationshipTo,
+    StringProperty,
+    StructuredNode,
+    StructuredRel,
+    config,
+    db,
+    install_all_labels,
+)
+
+# Some basic checks, and logging set up.
+while not str(pathlib.Path.cwd().name).lower().startswith("process-improve"):
+    os.chdir(pathlib.Path.cwd().parents[0])
+basecwd = pathlib.Path.cwd()
+sys.path.insert(0, str(basecwd))
+assert basecwd.exists()
+
+import process_improve.datasets.batch as batch_ds  # noqa: E402
+from process_improve.batch.data_input import melted_to_dict  # noqa: E402
+from process_improve.batch.plotting import plot_all_batches_per_tag  # noqa: E402
+
+pd.options.plotting.backend = "plotly"
+pd.options.display.max_columns = 20
+pd.options.display.width = 200
+config.AUTO_INSTALL_LABELS = True
+config.FORCE_TIMEZONE = True
+config.DATABASE_URL = "bolt://neo4j:@localhost:7687"
+
+
+# Code specific settings
+n_levels = 80
+percentile_low, percentile_high = 0.05, 0.95  # we create a range of n_levels between this low and high value
+
+# Ideally, use more than 1 tag to align on. These columns must exist in all data frames for all batches.
+columns_to_import = ["AgitatorPower", "AgitatorTorque", "JacketTemperature", "DryerTemp"]
+tag_to_plot = columns_to_import[2]
+
+
+# %%
+# Import the data: a dictionary of dataframes
+dryer_raw = (
+    pd.read_csv(pathlib.Path(batch_ds.__path__._recalculate()[0]) / "dryer.csv")
+    .rename({"ClockTime": "Pulse"}, axis=1)
+    .astype({"Pulse": "int"})
+)
+
+# %%
+# Set up the graph data model
+db.cypher_query("MATCH (n:Pitch) DETACH DELETE n;")
+
+
+class LeadsTo(StructuredRel):
+    """A relationship between two pitches."""
+
+    transition_duration = FloatProperty(required=True)
+
+
+class Pitch(StructuredNode):
+    """A pitch is a single data point in a batch."""
+
+    batch_id = StringProperty(required=True)
+    tag = StringProperty(required=True)
+    time_since_start = FloatProperty()
+    pulse = IntegerProperty()
+    datetime = DateTimeProperty()
+    pitch_value = IntegerProperty()
+    leads_to = RelationshipTo("Pitch", "LEADS_TO", model=LeadsTo)
+
+
+install_all_labels()
+
+
+# %%
+# Plot some data, to get an idea of what is present
+plot_all_batches_per_tag(
+    df_dict=melted_to_dict(dryer_raw, batch_id_col="batch_id"),
+    tag=tag_to_plot,
+    time_column="ClockTime",
+    x_axis_label="Time [hours]",
+)
+
+
+# %%
+# For all the batches, in one go, create a range of n_levels between the low and high percentile.
+for column in columns_to_import:
+    series = dryer_raw[column]
+    percentiles = series.quantile([percentile_low, percentile_high])
+
+    # Create a range of n_levels between the low and high percentile. then map the data to these levels,
+    # but don't clip the values; rather allow them to be outside the range, even if negative.
+
+    bins, step_size = np.linspace(percentiles[percentile_low], percentiles[percentile_high], n_levels + 1, retstep=True)
+    # Extend the bins to the left and right to allow for values outside the range, in the same step size
+    all_bins = np.concatenate(
+        [
+            np.flip(np.arange(start=bins[0], stop=series.min() - step_size, step=-step_size))[:-1],
+            bins,
+            np.arange(start=bins[-1], stop=series.max() + step_size, step=+step_size)[1:],
+        ]
+    )
+    # Overwrite the data with the levels
+    dryer_raw[column + "_level"] = pd.cut(series, bins=all_bins, labels=range(len(all_bins) - 1), include_lowest=True)
+
+
+# %%
+def store_row(row: pd.Series, batch_id: str, flags: pd.Series | None = None) -> dict:
+    """
+    Store a row of "pitches" representing batch data.
+
+    An optional flags series can be passed to indicate which tags to store.
+
+    """
+    pitches = {}
+    for tag_name, pitch_value in row.items():
+        tag = tag_name.replace("_level", "")
+        if tag == "Pulse":
+            continue
+        if (flags is not None) and (flags[tag_name] == 0):
+            continue
+        pitches[tag] = Pitch(
+            batch_id=batch_id,
+            tag=tag,
+            time_since_start=row["Pulse"] * 60 * 60,  # happens to be in hours; convert to seconds
+            pulse=int(row["Pulse"]),
+            pitch_value=int(pitch_value),
+        ).save()
+    return pitches
+
+
+# %%
+# Next steps: graph storage of the nodes, and then create a graph of the transitions between the nodes.
+# Do this per batch.
+df_dict = melted_to_dict(dryer_raw, batch_id_col="batch_id")
+
+for batch_id, df in df_dict.items():
+    this_batch = df[["Pulse", *[f"{item}_level" for item in columns_to_import]]]
+    # Store the first row
+    row_values = this_batch.iloc[0]
+    pitches = store_row(row_values, batch_id, flags=None)
+    # work with the diffs:
+    this_batch_diffs = this_batch.astype("int").diff()
+    for _, row in this_batch_diffs[1:].iterrows():  # we've handled row 1 already
+        if row.abs().sum() == 1:
+            continue  # we don't have any change from the prior row
+
+        row_values += row
+        # Create the transitions
+        for tag, pitch_value in (next_row := store_row(row_values, batch_id, flags=row)).items():
+            pitches[tag].leads_to.connect(
+                pitch_value, {"transition_duration": pitch_value.time_since_start - pitches[tag].time_since_start}
+            )
+
+        # Update the pitches
+        pitches |= next_row
+# %%
+a = 2
+
+# %%
diff --git a/requirements.txt b/requirements.txt
index f8e3b8a..e0eb005 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -11,6 +11,7 @@ numba
 seaborn
 pydantic
 tqdm
+neomodel
 
 # Notebook imports
 ipywidgets