diff --git a/CHANGELOG.md b/CHANGELOG.md
index 525db12..30b1c5a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,13 @@
 
 All notable changes to this project will be documented in this file.
 
+## [0.5.0] - 2023-04-26
+
+### Added
+
+- Added tasks `begin_sql_transaction` and `end_sql_transaction` to the `utils`module. These enable the management of SQL
+  transactions in flows. It also allows for dry running SQL statements.
+
 ## [0.4.0] - 2023-02-08
 
 ### Added
diff --git a/README.md b/README.md
index f5b5876..e3abcf9 100644
--- a/README.md
+++ b/README.md
@@ -120,6 +120,31 @@ with Flow(...) as flow:
     close_ssh_tunnel.run(tunnel=tunnel, upstream_tasks=[mysql_closed])
 ```
 
+**Use SQL transactions and dry running**
+
+```python
+from lolafect.connections import connect_to_mysql, close_mysql_connection
+from lolafect.utils import begin_sql_transaction, end_sql_transaction
+
+with Flow(...) as flow:
+    connection = connect_to_mysql(
+        mysql_credentials={...}, # You probably want to get this from TEST_LOLACONFIG.DW_CREDENTIALS
+    )
+    transaction_started = begin_sql_transaction(connection)
+    task_result = some_task_that_needs_mysql(
+        connection=connection,
+        upstream_task=[transaction_started]
+    )
+    transaction_finished = end_sql_transaction(
+        connection, 
+        dry_run=False, # True means rollback, False means commit changes
+        upstream_tasks=[task_result]
+    )
+    
+    close_mysql_connection(connection=connection, upstream_tasks=[transaction_finished])
+
+```
+
 ### Use Great Expectations
 
 **Run a Great Expectations validation on a MySQL query**
@@ -164,6 +189,13 @@ with Flow(...) as flow:
     )
 ```
 
+## Gallery
+
+This repo also contains a gallery of example flows that you can user to better
+understand `lolafect` or as templates to kickstart your own flows. You can 
+find these in `docs/gallery`.
+
+
 ## How to test
 
 There are two test suites: unit tests and integration tests. Integration tests are prepared to plug to some of our
diff --git a/docs/gallery/data_testing/README.md b/docs/gallery/data_testing/README.md
new file mode 100644
index 0000000..a1de4b0
--- /dev/null
+++ b/docs/gallery/data_testing/README.md
@@ -0,0 +1,7 @@
+# Data Testing Gallery
+
+In this folder, you can find a sample flow project that showcases how you can 
+do data testing with Lolafect.
+
+You can also take a look at our GE best practices and guidelines 
+[here](https://pdofonte.atlassian.net/wiki/spaces/DATA/pages/2484797445/Usage+Guidelines+and+Best+Practices).
\ No newline at end of file
diff --git a/docs/gallery/data_testing/data_testing_flow.py b/docs/gallery/data_testing/data_testing_flow.py
new file mode 100644
index 0000000..683237c
--- /dev/null
+++ b/docs/gallery/data_testing/data_testing_flow.py
@@ -0,0 +1,238 @@
+### INTRO
+
+# This is an example flow to showcase data testing.
+
+# There are several ways you can use this:
+# 1. If you simply want to copy paste useful recipes... the flow is yours.
+# 2. If you want to learn, I would advice:
+#     - Skim through the whole script.
+#     - Now read the flow block carefully and refer to other parts of the
+#       script when needed.
+#     - Try to run the flow as-is. It should succeed.
+#     - Try to intentionally break the data test by changing the data or the
+#       expectations.
+
+### HOW TO RUN
+# The flow is packed with comments to guide you through what's happening.
+# The flow is also runnable. To run it:
+# - Make a virtual environment with the requirements.txt that live in the same
+#   folder as this script.
+# - Start a shell, activate the venv, login to AWS and turn on your Mercadão
+#   VPN.
+# - In the shell, run the command: prefect run -p docs/gallery/data_testing/data_testing_flow.py
+#
+# Note: this flow is designed to run in your laptop. It won't work in the
+# prefect server. Don't bother uploading it.
+
+# The flow connects to DW and makes a silly check on a silly query. You can use
+# it as a reference on how to set up a data test in your serious flows.
+
+
+### IMPORTS
+
+import datetime
+
+from prefect import Flow, task, case
+from prefect.run_configs import KubernetesRun
+
+# ↑↑↑ Standard prefect stuff for the flow.
+
+from great_expectations.core.expectation_configuration import ExpectationConfiguration
+
+# ↑↑↑ ExpectationConfiguration is the class that allows us to define a single
+# expectation. We use it once for every expectation we define.
+
+from lolafect.lolaconfig import build_lolaconfig
+
+# ↑↑↑ Usual lolaconfig import to get all the env data.
+from lolafect.connections import (
+    open_ssh_tunnel_with_s3_pkey,  # ←←← We connect through an SSH tunnel
+    close_ssh_tunnel,  # ←←← Which we will have to close
+    connect_to_mysql,  # ←←← For quarantine purposes
+    close_mysql_connection,  # ←←← To close the previous connection
+)
+from lolafect.slack import SendSlackMessageTask
+
+# ↑↑↑ The task class to send slack messages.
+from lolafect.data_testing import (
+    run_data_test_on_mysql,
+)  # ←←← The task to run a data test
+
+### PREP
+
+LOLACONFIG = build_lolaconfig(flow_name="lolafect-gallery-data-testing-demo")
+# ↑↑↑ Get env from S3 and prepare everything related to it
+
+
+DATA_TEST_NAME = "gallery-example-test"
+# ↑↑↑ Our data test must have a name. We will need this if we want to look for
+# its logs in S3.
+
+DATA_TEST_QUERY = """
+    SELECT "hi" AS some_string,
+    1 AS some_number,
+    NULL as some_null
+"""
+# ↑↑↑ Our query defines what data do we want to test. This is a silly select
+# with hardcoded values because this is a demo, but in a real environment, you
+# most probably will want to have a common SELECT [...] FROM [...] query that
+# fetches the data your want to test.
+
+DATA_TEST_EXPECTATIONS = [
+    ExpectationConfiguration(
+        expectation_type="expect_column_values_to_match_like_pattern",
+        kwargs={"column": "some_string", "like_pattern": "%hi%"},
+    ),
+    ExpectationConfiguration(
+        expectation_type="expect_column_values_to_be_between",
+        kwargs={"column": "some_number", "min_value": 1, "max_value": 1},
+    ),
+    ExpectationConfiguration(
+        expectation_type="expect_column_values_to_be_null",
+        kwargs={"column": "some_null"},
+    ),
+]
+
+# ↑↑↑ Our expectations define what data should be like. Each expectation is
+# defined with a call to ExpectationConfiguration. You can check a reference
+# of available expectations and how to call them here:
+# https://legacy.docs.greatexpectations.io/en/latest/reference/glossary_of_expectations.html
+
+
+@task
+def fetch_tunnel_host_and_port(ssh_tunnel):
+    host = ssh_tunnel.local_bind_address[0]
+    port = ssh_tunnel.local_bind_address[1]
+
+    return host, port
+
+
+# ↑↑↑ A small helper function to get the host and the port where the SSH
+# tunnel is listening inside this host.
+
+
+@task
+def fail(exception, message):
+    raise exception(message)
+
+
+# ↑↑↑ A small helper function to cause a task failure with a custom message.
+
+
+@task
+def quarantine_failed_data(connection, query_to_get_quarantine_data):
+    cursor = connection.cursor()
+
+    cursor.execute(
+        f"""
+        CREATE TABLE quarantine.{LOLACONFIG.FLOW_NAME_UDCS}_{datetime.datetime.now().strftime("%Y%m%d_%H%M%S")} AS 
+        {query_to_get_quarantine_data}
+        """
+    )
+    # ↑↑↑ This query will store the faulty data in a quarantine schema in DW
+    # It creates a new table on each run, and uses the flow name and the current time
+    # to give the table a unique and informative name.
+
+
+send_slack_message = SendSlackMessageTask()
+# ↑↑↑ Simply making an instance of the task class. send_slack_message will be
+# the task we use in the flow.
+
+### FLOW
+
+with Flow(
+    LOLACONFIG.FLOW_NAME_UDCS,
+    storage=LOLACONFIG.STORAGE,
+    run_config=KubernetesRun(
+        labels=LOLACONFIG.KUBERNETES_LABELS,
+        image=LOLACONFIG.KUBERNETES_IMAGE,
+    ),
+) as flow:
+
+    ssh_tunnel = open_ssh_tunnel_with_s3_pkey(
+        s3_bucket_name=LOLACONFIG.S3_BUCKET_NAME,
+        ssh_tunnel_credentials=LOLACONFIG.SSH_TUNNEL_CREDENTIALS,
+        remote_target_host=LOLACONFIG.DW_CREDENTIALS["host"],
+        remote_target_port=LOLACONFIG.DW_CREDENTIALS["port"],
+    )
+    # ↑↑↑ We open an SSH tunnel pointing to DW
+
+    # ↓↓↓ This is where we actually run the data test. The result of the test
+    # gets stored in data_test_result.
+    data_test_result = run_data_test_on_mysql(
+        name=DATA_TEST_NAME,  # ←←← The name we set earlier
+        # ↓↓↓ The credentials to the MySQL where the data lives. We pass the
+        # ssh tunnel host and port instead of the true MySQL because we want
+        # to pass through the tunnel. If it was a direct connection, we would
+        # simply use the MySQL true host and port.
+        mysql_credentials={
+            "host": fetch_tunnel_host_and_port(ssh_tunnel)[0],
+            "port": fetch_tunnel_host_and_port(ssh_tunnel)[1],
+            "user": LOLACONFIG.DW_CREDENTIALS["user"],
+            "password": LOLACONFIG.DW_CREDENTIALS["password"],
+            "db": "sandbox",  # ←←← We always need to pass a default db, but it
+            # is recommended to always specify your schemas
+        },  # in the queries regardless.
+        query=DATA_TEST_QUERY,  # ←←← The query we set earlier
+        expectation_configurations=DATA_TEST_EXPECTATIONS,  # ←←← Idem
+        upstream_tasks=[ssh_tunnel],  # ←←← We must wait for the tunnel to be ready
+    )
+
+    # ↑↑↑ will take care of everything: connecting to S3 and DW, generate all
+    # the necessary configurations, run the actual test and store results both
+    # in memory and in S3.
+    #
+    # What to do from here is up to you. You can easily check if the test
+    # passed or not by accessing data_test_result["success"]. If it equals
+    # True, the test passed. If it equals False, at least one expectation
+    # failed.
+    #
+    # The following snippets are optional. You should judge if you want to do
+    # something similar or not in your flow based on your needs.
+
+    ### RAISING AN EXCEPTION
+    # When a test with run_data_test_on_mysql fails, it's important that you
+    # keep in mind that this will not cause a failure, in the sense of a
+    # prefect task failing. This is intentional: we didn't want to assume that
+    # a failing data test always translates into a failed flow.
+    #
+    # Nevertheless, it might be the case that you want your flow to fail if
+    # the data test didn't pass. To do so, you can use a simple helper task and
+    # a case, just like this:
+
+    with case(data_test_result["success"], False):
+        fail(ValueError, "Woops, the test didn't pass.")
+
+    ### SENDING A SLACK WARNING
+    # You might also want to send a slack message to a channel if the test
+    # does not pass. You can do so like this:
+
+    with case(data_test_result["success"], False):
+        send_slack_message(
+            LOLACONFIG.SLACK_WEBHOOKS["data-team-alerts-testing"],  # ←←← A webhook URL
+            "Uh oh, the demo flow failed.",  # ←←← Your warning message
+        )
+
+    ### QUARANTINE THE TESTED DATA
+    # Another common practice is to store the data that doesn't pass a test.
+    # This provides a lot of benefits that are discussed in our best practices
+    # docs in Confluence. Here is an example of you can quarantine the data
+    # that made your test fail:
+
+    dw_connection = connect_to_mysql(
+        mysql_credentials=LOLACONFIG.DW_CREDENTIALS,
+        overriding_host_and_port=fetch_tunnel_host_and_port(ssh_tunnel),
+        upstream_tasks=[data_test_result],
+    )
+    # ↑↑↑ We connect to DW, and since we are using the SSH tunnel, we
+    # override DWs host and port and instead use the listening ones from
+    # the tunnel.
+
+    with case(data_test_result["success"], False):
+        quarantined = quarantine_failed_data(dw_connection, DATA_TEST_QUERY)
+        # ↑↑↑ We call the quarantine_failed_data task. You can review the
+        # actions of this task in the definition that appears earlier in this
+        # file.
+
+    mysql_closed = close_mysql_connection(dw_connection)
+    tunnel_closed = close_ssh_tunnel(ssh_tunnel, upstream_tasks=[mysql_closed])
diff --git a/docs/gallery/data_testing/requirements.txt b/docs/gallery/data_testing/requirements.txt
new file mode 100644
index 0000000..245292b
--- /dev/null
+++ b/docs/gallery/data_testing/requirements.txt
@@ -0,0 +1,4 @@
+prefect==1.2.2
+great_expectations==0.15.45
+SQLAlchemy==1.4.46
+lolafect==0.4.0
\ No newline at end of file
diff --git a/lolafect/__version__.py b/lolafect/__version__.py
index 6a9beea..3d18726 100644
--- a/lolafect/__version__.py
+++ b/lolafect/__version__.py
@@ -1 +1 @@
-__version__ = "0.4.0"
+__version__ = "0.5.0"
diff --git a/lolafect/utils.py b/lolafect/utils.py
index 621e59d..2b7e460 100644
--- a/lolafect/utils.py
+++ b/lolafect/utils.py
@@ -1,5 +1,8 @@
 import json
+from typing import Any
 
+import prefect
+from prefect import task
 
 class S3FileReader:
     """
@@ -22,3 +25,41 @@ class S3FileReader:
             .read()
             .decode("utf-8")
         )
+
+@task()
+def begin_sql_transaction(connection: Any) -> None:
+    """
+    Start a SQL transaction in the passed connection. The task is agnostic to
+    the SQL engine being used. As long as the connection object implements a
+    begin() method, this task will work.
+
+    :param connection: the connection to some database.
+    :return: None
+    """
+    logger = prefect.context.get("logger")
+    logger.info(f"Starting SQL transaction with connection: {connection}.")
+    connection.begin()
+
+
+@task()
+def end_sql_transaction(connection: Any, dry_run: bool = False) -> None:
+    """
+    Finish a SQL transaction, either by rolling it back or by committing it.
+    The task is agnostic to the SQL engine being used. As long as the
+    connection object implements a `commit` and a `rollback` method, this task
+    will work.
+
+    :param connection: the connection to some database.
+    :param dry_run: a flag indicating if persistence is desired. If dry_run
+    is True, changes will be rolledback. Otherwise, they will be committed.
+    :return: None
+    """
+    logger = prefect.context.get("logger")
+    logger.info(f"Using connection: {connection}.")
+
+    if dry_run:
+        connection.rollback()
+        logger.info("Dry-run mode activated. Rolling back the transaction.")
+    else:
+        logger.info("Committing the transaction.")
+        connection.commit()
diff --git a/tests/test_integration/test_utils.py b/tests/test_integration/test_utils.py
new file mode 100644
index 0000000..1ce5f85
--- /dev/null
+++ b/tests/test_integration/test_utils.py
@@ -0,0 +1,123 @@
+import pytest
+
+from lolafect.lolaconfig import build_lolaconfig
+from lolafect.connections import (
+    open_ssh_tunnel_with_s3_pkey,
+    get_local_bind_address_from_ssh_tunnel,
+    close_ssh_tunnel,
+    connect_to_mysql,
+    close_mysql_connection,
+)
+from lolafect.utils import begin_sql_transaction, end_sql_transaction
+
+# __          __     _____  _   _ _____ _   _  _____ _
+# \ \        / /\   |  __ \| \ | |_   _| \ | |/ ____| |
+#  \ \  /\  / /  \  | |__) |  \| | | | |  \| | |  __| |
+#   \ \/  \/ / /\ \ |  _  /| . ` | | | | . ` | | |_ | |
+#    \  /\  / ____ \| | \ \| |\  |_| |_| |\  | |__| |_|
+#     \/  \/_/    \_\_|  \_\_| \_|_____|_| \_|\_____(_)
+# This testing suite requires:
+# - The calling shell to have permission in AWS
+# - The calling shell to be within the Mercadão network
+# - Do not use this tests as part of CI/CD pipelines since they are not idempotent and
+#   rely external resources. Instead, use them manually to check yourself that things
+#   are working properly.
+
+
+TEST_LOLACONFIG = build_lolaconfig(flow_name="testing-suite")
+
+
+@pytest.fixture
+def connection_with_test_table():
+    """
+    Connects to DW, creates a test table in the sandbox env, and yields the
+    connection to the test.
+
+    After the test, the table is dropped and the connection is closed.
+    """
+    test_local_bind_host = "127.0.0.1"
+    test_local_bind_port = 12345
+
+    tunnel = open_ssh_tunnel_with_s3_pkey.run(
+        s3_bucket_name=TEST_LOLACONFIG.S3_BUCKET_NAME,
+        ssh_tunnel_credentials=TEST_LOLACONFIG.SSH_TUNNEL_CREDENTIALS,
+        remote_target_host=TEST_LOLACONFIG.DW_CREDENTIALS["host"],
+        remote_target_port=TEST_LOLACONFIG.DW_CREDENTIALS["port"],
+        local_bind_host=test_local_bind_host,
+        local_bind_port=test_local_bind_port,
+    )
+
+    connection = connect_to_mysql.run(
+        mysql_credentials=TEST_LOLACONFIG.DW_CREDENTIALS,
+        overriding_host_and_port=get_local_bind_address_from_ssh_tunnel.run(
+            tunnel=tunnel
+        ),
+    )
+    cursor = connection.cursor()
+    cursor.execute("""
+        CREATE TABLE sandbox.lolafect_transaction_test_table
+        (
+            a_test_column INT
+        )
+        """)
+
+    # Connection and table ready for tests
+    yield connection  # Test happens now
+    # Test finished, time to remove stuff and close connection
+
+    cursor.execute("""
+        DROP TABLE sandbox.lolafect_transaction_test_table
+        """
+                   )
+    close_mysql_connection.run(connection=connection)
+    close_ssh_tunnel.run(tunnel=tunnel)
+
+
+def test_sql_transaction_persists_changes_properly(connection_with_test_table):
+    cursor = connection_with_test_table.cursor()
+
+    cursor.execute("""
+    SELECT a_test_column
+    FROM sandbox.lolafect_transaction_test_table
+    """)
+    table_is_empty_at_first = not bool(cursor.fetchall())  # An empty tuple yields False
+
+    begin_sql_transaction.run(connection=connection_with_test_table)
+    cursor.execute("""
+    INSERT INTO sandbox.lolafect_transaction_test_table (a_test_column)
+    VALUES (1)
+    """)
+    end_sql_transaction.run(connection=connection_with_test_table, dry_run=False)
+
+    cursor.execute("""
+    SELECT a_test_column
+    FROM sandbox.lolafect_transaction_test_table
+    """)
+    table_has_a_record_after_commit = bool(cursor.fetchall())  # A non-empty tuple yields True
+
+    assert table_is_empty_at_first and table_has_a_record_after_commit
+
+
+def test_sql_transaction_rollbacks_changes_properly(connection_with_test_table):
+    cursor = connection_with_test_table.cursor()
+
+    cursor.execute("""
+    SELECT a_test_column
+    FROM sandbox.lolafect_transaction_test_table
+    """)
+    table_is_empty_at_first = not bool(cursor.fetchall())  # An empty tuple yields False
+
+    begin_sql_transaction.run(connection=connection_with_test_table)
+    cursor.execute("""
+    INSERT INTO sandbox.lolafect_transaction_test_table (a_test_column)
+    VALUES (1)
+    """)
+    end_sql_transaction.run(connection=connection_with_test_table, dry_run=True)
+
+    cursor.execute("""
+    SELECT a_test_column
+    FROM sandbox.lolafect_transaction_test_table
+    """)
+    table_is_still_empty_after_rollback = not bool(cursor.fetchall())  # A tuple yields False
+
+    assert table_is_empty_at_first and table_is_still_empty_after_rollback