meridian-oss 0.2.0

Heroku for ML Features

Meridian

Heroku for ML Features

"Define features in Python. Get training data and production serving for free."

Meridian is a developer-first feature store designed to take you from a Jupyter notebook to production in 30 seconds. It eliminates the infrastructure tax of existing tools—no Kubernetes, no Spark, no YAML. Just pure Python and SQL.

---

⚡ The 30-Second Quickstart

1. Install

pip install "meridian-oss[ui]"

2. Try the Quickstart Script Run a self-contained script to see the API in action:

python examples/quickstart.py

3. Define Features (examples/basic_features.py) Or define features in a file to serve them:

from meridian.core import FeatureStore, entity, feature
from datetime import timedelta

store = FeatureStore()

@entity(store)
class User:
    user_id: str

@feature(entity=User, refresh=timedelta(minutes=5), materialize=True)
def user_click_count(user_id: str) -> int:
    return 42

4. Serve

meridian serve examples/basic_features.py

5. Query

curl -X POST http://localhost:8000/features \
  -H "Content-Type: application/json" \
  -d '{"entity_name": "User", "entity_id": "u1", "features": ["user_click_count"]}'

# {"user_click_count": 100}

6. Visualize Launch the built-in UI to explore your features interactively:

meridian ui examples/basic_features.py

Or see the rich terminal dashboard:

meridian serve examples/basic_features.py

---

🚀 Why Meridian?

1. Local-First, Cloud-Ready

Most feature stores require a platform team to set up. Meridian runs on your laptop with zero dependencies (DuckDB + In-Memory) and scales to production with boring technology (Postgres + Redis).

2. No Magic, Just Python

We don't use YAML for configuration. Your code is your config.

• Explicit Caching: Use @feature(materialize=True) to cache hot features.
• Explicit Refresh: Use @feature(refresh="5m") to define freshness.

3. Instant Wow ✨

• Meridian UI: A built-in Streamlit dashboard to explore your data.
• Rich Terminal: Production-grade TUI with live metrics.
• Jupyter Integration: Beautiful HTML representations of your feature store objects.

4. Production-Grade Reliability

• Redis-Only Caching: We avoid complex multi-tier caches to guarantee data consistency.
• Randomized Locking: Our distributed scheduler is self-healing and requires no leader election.
• Zero-Code Serving: Auto-generated FastAPI endpoints with built-in metrics and logging.

---

🏗️ Architecture

Tier 1: Local Development (The "Wedge")

Perfect for prototyping and single-developer projects.

• Offline Store: DuckDB (Embedded)
• Online Store: Python Dictionary (In-Memory)
• Scheduler: APScheduler (Background Thread)
• Infrastructure: None (Just pip install)

Tier 2: Production (The "Standard")

Robust, scalable, and boring.

• Offline Store: Postgres / Snowflake / BigQuery
• Online Store: Redis
• Scheduler: Distributed Workers with Redis Locks
• Infrastructure: 1x Postgres, 1x Redis, Nx API Pods

---

🗺️ Roadmap

• Phase 1 (Now): Core API, DuckDB/Postgres support, Redis caching, FastAPI serving.
• Phase 2: Drift detection, RBAC, and multi-region support.

---

🤝 Contributing

We love contributions! Please read our CONTRIBUTING.md to get started.

📄 License

Apache 2.0