Private AI guidance tool

Brand Guidance Chatbot

Krystian built a brand-guidance prototype with a source-update path, sensitive-information guardrails, user testing, and escalation for out-of-scope questions.

Internal Systems Prototype 2024 - Present
QuestionApproved brand sourcesAnswer with boundariesEscalate when uncertain
Prototype retrieval path I designed for internal brand guidance, with source updates, sensitive-information boundaries, and escalation.

Exact role

Senior Web Developer (AI workflow prototyping)

University of Arizona

Current status

Prototype

2024 - Present

Problem

What needed to change

Campus marketers needed faster access to brand guidance so they could answer policy and copy questions without manually searching the MarCom site or asking the same support questions repeatedly.

Users

Who the system serves

University marketers, campus communicators, and internal teams responsible for brand-aligned messaging.

Team context

Where I fit

Owned the prototype and led internal testing with team members and intended campus users.

Important technical decision

Build a source update path and refuse or escalate sensitive and out-of-scope questions.

Tradeoff or limitation

The prototype is useful for finding guidance but remains in user testing; it is not presented as a campus-wide production system.

Implementation

What I personally owned

  • Owned the project from inception to prototype.
  • Led internal testing with team members and intended campus users.
  • Designed the source update, refusal, and escalation paths.

Constraints

Hard parts that shaped the work

  • Designing an update process so the chatbot could rebuild when the source site changed.
  • Adding safeguards so the bot would not collect sensitive information or answer questions it should not answer.
  • Balancing helpful guidance with trust, boundaries, and policy-sensitive escalation.

Prototype output

What exists now

  • Working prototype for brand and policy guidance.
  • Prototype source update and rebuild process.
  • Guardrails for sensitive information and out-of-scope answers.
  • Internal testing workflow with team members and campus users.

Result

Verified result and current state

  • Team members use the prototype as an alternative to browsing the MarCom site for information.
  • It remains a prototype in user testing; broader campus adoption is not established.

Technical scope

Tools used

LLM workflowsBrand and policy guidanceRetrieval workflow prototypeChatbot testingContent governance