Supercharge ML Data Prep in a Few Lines of Python

Generate ML batches directly from your data warehouse or data lake. Clean, transform, sample, batch and more using Tempora's powerful data preparation technology.

Less Munging, More Modeling

Data scientists and ML engineers spend most of their time wrangling raw data. Tempora makes it easy to build powerful ML data prep workflows, freeing your team to focus on modeling and insights.

ML That Respects Data Governance

No more SQL exports to generate datasets for ML. Tempora integrates natively with where your data lives, preserving governance, maintaining lineage and eliminating data duplication.

Runs on Your Infra

With Tempora, your data never leaves your environment. Purpose-built for self-hosting, Tempora Server is fully containerized and runs effortlessly in your VPC or on-premises.

Connects directly to your data sources

Experience data preparation without the pain – no more exports, duplication, or broken lineage.

How Tempora works

$ docker compose up -d
[+] Running 2/2

✔ Network tempora-server_default    Created
✔ Container tempora-server          Started

$ docker compose logs -f
tempora-server  | 2025-10-20 09:45:31 INFO: Starting server...
tempora-server  | 2025-10-20 09:45:31 INFO: Started server on port 7000

    
from tempora.datasets import FSDataset, Pivot

sensor_data = FSDataset(
    source='s3://tempora-ai/datasets/industrial/sensor_data',
    partitioning='hive', time_column='datetime',
    entity_keys=['Machine Serial'],
    pivot=Pivot(
        on='param', using='value',
        agg_function='first', dtype='float64',
    ),
)

machine_data = FSDataset(
    source='~/industrial/machine_data.csv',
)

claims_data = FSDataset(
    source='~/industrial/claims_data.json',
    time_column='Failure Date',
)

    
dataset = sensor_data.join(
    machine_data,
    join_condition=['Machine Serial'],
)

dataset = dataset.join(
    claims_data,
    join_condition='"Machine Serial" = "Product Serial Number"',
    asof_join=True,
)

dataset = dataset.filter(
    "Model = 'XC-500' AND ENGINE_HOURS IS NOT NULL",
)

    
from tempora.samplers import RandomSampler, EventPrediction, Length

context_len = Length(pd.Timedelta('5 hours'))
target_len = Length(pd.Timedelta('20 hours'))

event_target_gen = EventPrediction(
    window_len=target_len,
    columns=['Failure Date'],
)

sampler = RandomSampler(
    context_len=context_len,
    batch_size=32,
    target_gen=event_target_gen,
    output_format='np',
)

for batch in sampler(dataset):
    ...