Jeff Hajewski

Reliable AI systems, evals, context, and performance economics.

← Back to projects

Nori

A distributed key-value store built to explore the mechanics of consistency, replication, and operability.

Creator · Research project

Why I built it

I like building systems that expose the real tradeoffs instead of hiding them behind abstractions. Nori is a way to work through the operational surface area of a distributed database from first principles: replication, storage, failure detection, sharding, and the ergonomics required to debug it.

What it does

Nori is a distributed key-value store written in Rust for high availability and strong consistency. It combines Raft consensus for replication, an LSM-tree storage engine with leveled compaction, Jump Consistent Hashing for sharding, and SWIM for gossip-based failure detection.

It ships with client SDKs for TypeScript, Python, Go, and Java, along with built-in observability: metrics, tracing, and live cluster visualization.

Why it matters

Even when my day-to-day work is closer to AI infrastructure, the underlying engineering problems still rhyme with distributed systems: keeping state legible, making failures observable, and building enough mechanical sympathy into the architecture that the system behaves under pressure.