Mubayyin Release Intelligence

An AI release-assurance agent that investigates whether an AI system is ready to serve customers by querying structured release records, test results, policies, findings, and evidence stored in Sanity.

What I Built

Mubayyin Release Intelligence helps teams investigate AI releases before they reach customers.

Instead of asking an AI assistant to give a generic opinion about whether an AI system is ready, the agent queries a structured release knowledge base containing:

• AI systems and versions
• Models and prompt versions
• Release records
• Test suites and test cases
• Arabic and English scenarios
• Findings and failures
• Evidence records
• Policies
• Previous release decisions
• Regression history

The agent can answer questions such as:

Can this release be served to customers?

Why was this release held?

What changed between two releases?

Which tests failed?

What evidence supports the finding?

Which policy applies to the failure?

The goal is to demonstrate how structured content can give an agent the context needed to perform an evidence-based investigation rather than simply generate a generic answer.

Demo

[ADD DEMO VIDEO LINK]

Code

[ADD PUBLIC GITHUB REPOSITORY]

How I Used Sanity

Sanity is the structured knowledge layer behind the agent.

I modeled AI release information as connected content rather than one large document. Releases reference their AI system, test suite, policies, findings, decisions, and supporting evidence.

The agent uses Sanity Context to query this structured content while investigating a release.

For example, when asked whether version 1.4 can be released, the agent retrieves the release record, follows its related test results and findings, retrieves the supporting evidence, and checks the applicable policy before explaining the stored release decision.

This makes the relationships between the content important to the agent's answer.

The same investigation could then be continued with questions such as:

• "What changed from version 1.3?"
• "Which failures remain unresolved?"
• "Show me the evidence for the escalation failure."
• "Which policy applies to this finding?"

Sanity therefore acts as more than a content store: it provides the structured context the agent uses to investigate the release.

Sanity Project Details

Sanity Project ID:

[ADD PROJECT ID]

Dataset:

[ADD DATASET]

Public dataset/project link:

[ADD LINK IF AVAILABLE]

Agent Session

[ADD PUBLIC AGENT SESSION LINK]

The session demonstrates the agent querying the Sanity content and using the retrieved release records, findings, policies, and evidence to answer a release-readiness investigation.

Why This Matters

AI systems increasingly change through model updates, prompts, retrieval configuration, policies, and other components.

Mubayyin Release Intelligence explores a simple question:

Can an agent investigate the evidence surrounding an AI release instead of relying only on a model's general knowledge?

The prototype demonstrates that workflow using structured content in Sanity.