Does TensorFlow help my target role?
☐ Job posts mention ML pipelines, feature stores or model training data
☐ The company trains deep learning models in-house
☐ The team uses TensorFlow or Keras
☐ I already know SQL, Python and one orchestration tool

3-4 ticks → learn TensorFlow's data side now
1-2 ticks → strengthen core DE skills first

python -m venv tf-env
source tf-env/bin/activate    # Windows: tf-env\Scripts\activate
pip install tensorflow pandas
python -c "import tensorflow as tf; print(tf.__version__)"

import pandas as pd
import tensorflow as tf

df = pd.DataFrame({"age": [25, 32, 47, 51], "income": [3.2, 4.8, 6.1, 7.4],
                   "label": [0, 0, 1, 1]})
features = df[["age", "income"]].astype("float32").values
labels = df["label"].values

ds = (tf.data.Dataset.from_tensor_slices((features, labels))
        .shuffle(buffer_size=len(df))
        .batch(2)
        .prefetch(tf.data.AUTOTUNE))

for x, y in ds.take(1):
    print(x.shape, y.numpy())

import tensorflow as tf

mean = tf.constant([38.75, 5.375])
std = tf.constant([10.0, 1.5])

def normalise(x, y):
    return (x - mean) / std, y

ds = ds.map(normalise, num_parallel_calls=tf.data.AUTOTUNE)
# normalisation now happens inside the pipeline,
# so training and inference can share the same logic

import tensorflow as tf

def make_example(age, income, label):
    return tf.train.Example(features=tf.train.Features(feature={
        "age": tf.train.Feature(float_list=tf.train.FloatList(value=[age])),
        "income": tf.train.Feature(float_list=tf.train.FloatList(value=[income])),
        "label": tf.train.Feature(int64_list=tf.train.Int64List(value=[label])),
    }))

rows = [(25.0, 3.2, 0), (47.0, 6.1, 1)]
with tf.io.TFRecordWriter("train.tfrecord") as w:
    for r in rows:
        w.write(make_example(*r).SerializeToString())

import tensorflow as tf

spec = {
    "age": tf.io.FixedLenFeature([], tf.float32),
    "income": tf.io.FixedLenFeature([], tf.float32),
    "label": tf.io.FixedLenFeature([], tf.int64),
}

def parse(record):
    ex = tf.io.parse_single_example(record, spec)
    return tf.stack([ex["age"], ex["income"]]), ex["label"]

ds = tf.data.TFRecordDataset("train.tfrecord").map(parse).batch(32)
for x, y in ds.take(1):
    print(x.numpy(), y.numpy())

Project: Churn Model Data Pipeline
Source: public customer dataset (CSV)
Pipeline steps:
  1. Ingest raw CSV into a staging table (SQLite / Postgres)
  2. Clean and build features with SQL + pandas
  3. Validate: no nulls in key columns, value ranges
  4. Export train/validation split to TFRecord
  5. tf.data pipeline with map, shuffle, batch, prefetch
  6. Tiny Keras model to prove the data works
Orchestration (optional): Airflow or a scheduled script

Data Validation Checklist
☐ Row count matches source (± expected drops)
☐ No nulls in label column
☐ Numeric columns inside expected ranges
☐ Train/validation split has no overlapping IDs
☐ Class balance recorded
☐ TFRecord read-back test passes
☐ Schema documented in README

# ML Data Pipeline
## Problem
## Architecture (source → staging → features → TFRecord → tf.data)
## How to run
## Data validation rules
## Performance notes (prefetch, parallel map)
## What I would add next (feature store, monitoring)