Vahdettin Karataş
Data & ML engineering — batch pipelines, APIs, validation
  • Location:
    Prague, Czech Republic
Technical focus
  • Batch feature pipelines
  • Definition registries & transforms
  • FastAPI & OpenAPI
  • pytest & CI
  • Reproducible data contracts

Interactive pipeline demo

Feature Store Mini — batch workflow with one code path for CLI, API, and notebooks.

Upload a CSV or run the bundled Telco-style sample. You get the same feature table and validation summary as the repository and the live service—useful when reviewers want evidence, not slides.

  • 8derived features
  • 1pipeline entrypoint
  • 4validation checks

How it works

  1. Definitions — each feature documents inputs, dtype, and meaning in one registry.
  2. Transform — a single build step produces customer_id + feature columns.
  3. Validate — duplicates, schema, and sanity checks before you trust the file.

Deliberately small: not Feast or Tecton — a reusable pattern for consistent offline features and a clear proof-of-work story.

Run the pipeline

Upload your own churn-style CSV, or run the bundled synthetic sample — no local file required. Same engine as POST /demo/transform on the API host.

Live on the API host

Upload a raw .csv (max 2 MB), or use Try bundled sample to fetch sample_raw.csv from the service (~250 synthetic rows).

The sample button calls the deployed API (GET /demo/sample-raw.csv) then POST /demo/transform — works from this static page when the API allows cross-origin requests.

Schema: churn-style columns (e.g. tenure, Contract, services).

Local CLI

After pip install -r requirements.txt:

python -m src.pipeline.build_feature_table

Default in: data/raw/sample_raw.csv → out: artifacts/feature_table/customer_feature_table.csv

Stricter raw checks: python -m src.pipeline.build_feature_table --strict

Results

JSON summaries below; preview is the first rows of the built feature table.

Meta


  

Input summary


  

Output summary


  

Validation


  

Feature columns


  

Preview rows

Feature Store Mini

Batch pipeline · definition registry · validation · FastAPI

© Vahdettin Karataş. All rights reserved.