Case study · Distributed systems · 2022–2026
Zircuit
The system that schedules and runs proof jobs for an EVM-compatible ZK-rollup, plus its release engineering, from the first commits to public launch.
- Backend and distributed systems
- Releases and reliability
- Early-stage teams
- Blockchain infrastructure

- Company
- Zircuit
- Product
- EVM-compatible ZK-rollup
- Timeline
- Sep 2022 – Feb 2026
- My role
- Senior backend engineer
- Stack
- Go · NATS · Rust · Optimism Bedrock
- Outcome
- Public mainnet, Aug 2024
On this page
Architecture
Challenge
Zircuit is an EVM-compatible ZK-rollup with AI-powered security at the sequencer. Every batch has to be proved, and if proving falls behind, state can’t advance and blocks can’t finalize. The proving pipeline had to keep up, survive prover upgrades, and never hold up the chain.
Constraints
- Proof jobs are long and expensive. A batch was proved every 5 to 6 hours, scaling to 8 to 10 provers under load.
- Every upgrade meant letting running jobs finish first, across staging, testing and production, with DevOps and external partners.
- Internal code had to stay private while the protocol code was published openly.
What I owned
I was among the first four engineers, from the first commits to public launch.
- Proof jobs. Architected the system that schedules and runs proof jobs, with a NATS message bus, Go consumers running jobs concurrently, and a Rust prover as the worker.
- Prover migrations. Moved the proving stack from Halo2 to SP1, helped move the pipeline onto a new job-scheduling system, and owned prover upgrades in production.
- Sequencer screening. Built detectors on the sequencer-level security interface, including the blocklist and allowlist that decide which transactions are admitted.
- Public release pipeline. Automated publishing of the open-source monorepo from the private tree, stripping internal code before each push.
- Release reliability. Let running jobs finish before each upgrade, restarted the pipeline after, and ran the rollback when one failed.
Approach
Never let proving hold up the chain. A durable message bus keeps producing batches separate from proving them, proof jobs run concurrently, and the pool of provers scales with load.
Upgrades are treated as operations. Let running jobs finish, upgrade, check it, restart, and keep a rollback ready.
Results
- Zircuit’s Mainnet Phase 1 went live on August 5, 2024.
- Publishing to the open-source monorepo became about 2.5× faster, with internal code stripped automatically.
- Proving scaled to 8 to 10 provers under load.
Zircuit has since moved its L2 to Conduit, and the public repositories are kept for legacy purposes.
Skills this shows
| If your role needs | Evidence in this project |
|---|---|
| Distributed systems | A NATS-based system that schedules and runs long proof jobs, scaled to 8–10 workers. |
| Release engineering | Upgrades that let running jobs finish, then restart, across three environments, with rollback. |
| Production migrations | Moved the prover from Halo2 to SP1 in production. |
| Security-sensitive systems | Transaction-screening detectors at the sequencer. |
| Open-source operations | An automated public mirror of a private monorepo. |
What this means for your project
- Running long, expensive jobs that can’t fall behind? I design the concurrency so a slow step queues work instead of holding everything else up.
- Upgrading critical infrastructure in production? I treat upgrades as operations. I let running jobs finish, upgrade, check it, restart, and keep a rehearsed rollback ready.
- Early team, no process yet? I’ve built release management and incident response from scratch while shipping features.
On the record
Zircuit’s own look back at the road from testnet to mainnet, and what comes next.
Zircuit’s announcement of its public mainnet launch.
Independent coverage of the mainnet launch.
The open-source monorepo, published by the release pipeline I built.