Private AI governance workflow

Logo Proposal Validation Agent

Krystian designed and shipped an internal logo-review workflow that moves clear cases in under one minute while keeping ambiguous decisions with reviewers.

Internal Systems Shipped internally 2024 - Present
ProposalPolicy checksRetrieved guidanceHuman decision
ApproveDenyClarifyEscalate
Internal logo-review workflow I designed. Deterministic checks and retrieved policy context support the reviewer; ambiguous cases stay in human review.

Exact role

Senior Web Developer (AI product architecture and technical lead)

University of Arizona

Current status

Shipped internally

2024 - Present

Problem

What needed to change

Logo and brand proposal approvals were consuming reviewer time because submissions needed policy checks, judgment, drafted replies, and follow-up for unclear cases.

Users

Who the system serves

Brand reviewers, campus submitters, and marketing stakeholders.

Team context

Where I fit

Krystian owned inception, architecture, development, publishing, documentation, and coordination with brand reviewers, marketing stakeholders, submitters, and development partners.

Important technical decision

Combine deterministic policy checks with retrieved guidance and structured LLM output, then route unclear or sensitive cases to a reviewer.

Tradeoff or limitation

Speed is valuable only for clear cases. Ambiguous requests stay human-in-the-loop rather than forcing an automated answer.

Implementation

What I personally owned

  • Owned project inception, architecture, development, publishing, documentation, and stakeholder coordination.
  • Designed the AI review flow from submission intake through approve, deny, ambiguous review request, and clarifying follow-up.
  • Coordinated across development and stakeholder groups as interest grew beyond the original reviewer.

Constraints

Hard parts that shaped the work

  • Retrieval quality depended on the quality and structure of the policy source data.
  • Discovered the request form itself had unclear fields and one question people often answered strategically just to move through the form faster.
  • The workflow needed to improve routine review without automating ambiguous or policy-sensitive decisions.

Shipped result

What exists now

  • AI review agent with approve, deny, ambiguous, and clarification paths.
  • RAG and deterministic-policy checks with structured outputs.
  • Channel notifications with approval and denial choices.
  • Clarifying-question flow for ambiguous submissions.
  • Tool calling to aid decisions as the system matured.

Result

Verified result and current state

  • Clear cases can move through the workflow in under one minute.
  • Ambiguous and policy-sensitive submissions remain with human reviewers.
Typical reviewer volume: roughly six submissions per day.Manual review, reply drafting, and outreach previously took about 10 minutes per submission.Clear cases can now move through the workflow in under a minute.

Technical scope

Tools used

PythonFastAPIOpenAI APIRAGEmbeddingspgvectorPostgresStructured JSONDeterministic rulesHuman reviewEscalation paths