# Design a model registry: system design interview question

> Large model artifacts, approval rules, concurrent release changes, and lineage that survives artifact deletion.

Source: https://sysround.com/questions/design-model-registry
Updated: 2026-11-05


Difficulty: hard. Levels: Senior, Staff+.

## How the interviewer opens it

"Design a model registry for teams that store model artifacts and manage approval and release. Preserve lineage and support safe promotion and rollback."

## What to clarify first

- Users and permissions
- Request volume
- Artifact volume
- Version identity and lineage
- Approval rules
- Promotion and rollback
- Latency
- Consistency and durability
- Retention
- Scope boundaries

## Scope of a strong answer

Core requirements:

- Register model versions and retrieve their artifacts
- Record each version's code, dataset, and training run references
- Approve versions for release
- Promote and roll back releases without bypassing approval rules

Usually out of scope:

- Train models or run evaluations
- Serve predictions or control deployment traffic
- Store source datasets
- Build a model security scanner

## What interviewers probe

### Artifact publication

Large uploads can stop midway. A published version must retain exact bytes despite retries, later uploads, and a zone failure.

### Approval decisions

Approval must identify exact content and an authorized reviewer. Concurrent revocation and release requests must produce a clear result.

### Release changes

Concurrent promotions and retries can select the wrong version. A zone failure must not erase an acknowledged release or its audit record.

### Lineage and retention

Artifact deletion must preserve release safety and useful lineage. External dataset references can disappear even when registry records remain.

## Common mistakes

- Treating a model version as a mutable directory
- Using one status field for artifact readiness, approval, and release
- Checking approval separately from the release change
- Assuming high availability guarantees durability without defining cross-zone acknowledgment rules
- Claiming that a registry rollback also restores deployment traffic
- Deleting artifacts without checking retained releases

## The bar by level

- **Senior:** Handles partial publication, request retries, concurrent release changes, and approval revocation. Defines cross-zone acknowledgment requirements for metadata and artifacts. Connects retention to rollback safety. Separates registry state from deployment state.
- **Staff+:** Defines transaction boundaries across approval, release, audit, and deletion. Explains quorum behavior and safe failover after a zone failure. Establishes policy for rule changes and external lineage loss. Quantifies artifact cost and regional failure trade-offs.

Practice this question with an AI interviewer: https://sysround.com/practice/new?problem=model-registry


---
SysRound (https://sysround.com) is an AI practice tool for system design interviews. It has a voice or typed AI interviewer and a shared whiteboard. At the end, you get feedback on each skill and a model answer. The AI feedback can be wrong.