← All Projects
AI UX · Product Concept Case Study

AI Compass

Helping people think with AI — not just use it. A guided decision-making experience designed to turn vague goals into confident, actionable direction.

Why this mattered: AI tools are powerful but directionless by default. Most users don't have a prompting problem — they have a clarity problem. AI Compass was built to address the gap between having access to AI and knowing what to actually do with it.
Role Product Designer / UX Strategist
Type Personal Product Concept
Focus AI UX · Decision Design · Product Strategy
Project Snapshot
Problem Most AI tools optimize for response speed, not decision quality — leaving users with outputs they can't evaluate or act on.
My Role Product Designer / UX Strategist — product concept, UX architecture, decision loop design, interaction model, information flows.
Key Users Anyone using AI without a clear framework — job seekers, product builders, researchers, professionals learning new AI workflows.
Constraints Guided without being prescriptive. Structured without feeling like a form. Helpful without removing user agency or creating dependency.
Key Contribution Decision loop architecture, guided intake model, assumption visibility layer, tradeoff comparison design, session artifact system.
Outcome A product concept that repositions AI from answer machine to thinking partner — structured around the human need for direction, not just output.

Most people use AI the way they used early search engines — hoping the query will somehow do the thinking.

The result is scattered chat threads with no durable structure, responses accepted without validation, and outputs that feel vague rather than useful. Users get something back — but they're not sure whether it's right, whether they asked the right question, or what to actually do with it now.

The problem is not that AI can't help. It's that the interface provides no guidance on what to ask, no structure for evaluating what comes back, and no path from response to decision. The blank prompt field is the most honest symbol of this gap: powerful technology, zero direction.

The core tension: AI adoption is rising, but the ability to use AI well is not keeping pace. Access to the tool is not the same as knowing how to use it.

What users experience

Blank prompt anxiety Unclear goals before starting Overtrusting AI responses No way to validate outputs Can't convert responses to action Reactive use, not strategic

What the tools fail to provide

Goal clarification before generation Structured option comparison Assumption transparency A path from response to next step Durable session artifacts

The next wave of AI products shouldn't just be more capable. They should make users more capable.

AI adoption is widespread. AI literacy is not keeping pace. Most people don't need another AI tool — they need a better way to think through what they're trying to do before they involve AI at all. The interface is the guidance. If the interface doesn't help users clarify their goals, they'll keep getting outputs that don't land.

AI Compass was built around the belief that the most valuable thing an AI product can do isn't give a faster answer — it's help the user ask a better question. That's a different design brief than most AI products are working from.

"The next wave of AI products shouldn't just be more capable. They should make users more capable. That's the design gap AI Compass was built around."

Good AI UX reduces cognitive fog — it doesn't add another dashboard. The best AI product for most users isn't the most powerful one. It's the one that helps them understand what they're actually trying to solve.

Defined less by who they are — more by where they're stuck.

AI Compass isn't for a specific demographic. It's for a specific user state — the moment when someone knows they want to use AI but doesn't know where to start, or has gotten an AI response and doesn't know what to do with it. That moment is more universal than any persona description can capture.

"I have an idea but I don't know where to start."
"I got an AI response but I'm not sure if it's right."
"I need help thinking through my options before I decide."
"I need a system, not another endless chat thread."
Job seekers positioning themselves Product builders shaping vague ideas Students organizing complex research Professionals building AI workflows Non-technical users who want guided support Creators turning ideas into plans

A decision loop — not a chat thread. Seven steps from vague goal to confident direction.

The core product concept is a structured loop that guides users through a decision instead of dropping them into a prompt field. Each step builds on the last. The system holds context so users don't have to re-explain themselves at every turn.

Clarify the goal

Define what you're actually trying to figure out before anything else happens. The system doesn't let users skip this step — vague input produces vague output.

Understand the context

Surface constraints, prior knowledge, and background that shapes the problem. Guided intake questions pull out what the user knows and what they don't.

Generate possible paths

Explore multiple approaches instead of defaulting to the first answer. Three structured options force comparison and prevent passive acceptance of the obvious path.

Evaluate tradeoffs

Compare effort, risk, and outcome across each option before choosing. The system surfaces the assumptions baked into each path so the user can challenge them.

Choose a direction

Select a path with documented reasoning you can return to later. The decision and the rationale behind it are captured together — not just the outcome.

Turn it into action

Convert the decision into next steps, prompts, or a working plan. The session ends with something the user can actually do — not just something they understand.

Save the learning

Archive useful decisions and frameworks as reusable artifacts. The product gets more valuable over time as the user's decision history accumulates.

Six passes — from problem framing to interaction model.

Identify the real barrier

Started by separating the access problem (do users have AI tools?) from the literacy problem (can users use them well?). Found that access was broadly solved; direction and confidence were not.

Map current AI user experience

Documented what actually happens when users open an AI tool — the blank prompt, the trial-and-error prompting, the unvalidated response, the stuck user who doesn't know if the answer is good. Identified each moment where direction was missing.

Design the decision loop

Built the seven-step framework from first principles: what does a good decision process look like, and how much of it can the interface support without becoming a burden? Each step was evaluated against one question: does this help the user, or does it help the system feel complete?

Define the interaction model

Designed guided intake to feel like a structured conversation, not a form. Tested the line between "enough structure to be helpful" and "so much structure it feels like homework." Progressive disclosure was the main tool for managing this balance.

Build the prompt scaffolding

Designed the system that translates user input into structured, context-aware prompts. The goal was that users never need to write a prompt — the intake process builds it for them from what they said.

Design for trust

Added the assumption visibility layer — surfacing what the AI is taking for granted before the user moves forward. Designed the tradeoff comparison view so options feel genuinely comparable, not artificially different. Both were about giving users the tools to evaluate, not just receive.

Four decisions that defined how the product thinks about user agency.

Guided intake over open prompt
Decision Start with structured intake questions before generating anything — not a blank prompt field.
Reason Users who don't know what to ask can't get useful answers from a blank field. The problem precedes the interface.
Tradeoff Slower time to first AI response. Some experienced users may find the intake unnecessary.
Result A higher quality starting point for the entire session — the intake pays for itself in output quality.
Multiple pathways, not one answer
Decision Show three structured options rather than one recommended output after the intake.
Reason A single recommendation invites passive acceptance. Multiple options force comparison and create a sense of agency.
Tradeoff More cognitive load upfront. Users who want a fast answer may feel slowed down.
Result Users who feel they chose — not just received. The decision belongs to them, not the AI.
Assumption visibility
Decision Surface what the AI is assuming about the user's context before surfacing the main options.
Reason Hidden assumptions cause users to trust outputs they shouldn't — and there's no way to course-correct if you can't see what the model assumed.
Tradeoff Adds a review step before the main content appears. Some users will skip it without reading.
Result Users who engage with assumptions can correct the model before the wrong direction compounds across the session.
Session artifacts
Decision Save the decision, reasoning, and options considered as a structured artifact at the end of each session.
Reason Useful thinking gets lost in scroll history. The value of a good decision shouldn't disappear when the chat window closes.
Tradeoff Requires session state management and a storage model. Adds product complexity.
Result Decisions become reusable references. The product accumulates value over time instead of resetting each session.

Four principles behind every interface decision. Six features that deliver them.

The interface needed to feel calm and intelligent — not like a productivity app with too many moving parts, and not like a chatbot pretending to be something smarter. The design principles drove every specific feature decision.

UX principles

Direction over answers. The product's job is to help users move toward clarity — not just produce responses.
Structure without friction. The system guides thinking without making the experience feel like filling out a form.
Transparency builds trust. Users should understand what the AI is assuming before acting on its suggestions.
Action beats output. The final value is not the AI response — it's what the user can do next.

Features that deliver them

Guided Prompt Builder Decision Pathways Assumption Check Tradeoff Comparison Action Plan Generator Session Memory

Each feature targets a specific failure mode in how users currently interact with AI. None of them are about making the AI more powerful — they're about making the user more capable.

"The best AI product for most users isn't the most powerful one. It's the one that helps them ask better questions."

AI Compass is not just screens. It's an information architecture.

The visible interface is the surface layer. Below it, a structured information flow determines how goals get classified, how options are scaffolded, how responses are evaluated, and how decisions become reusable. This is where my background in UX, data systems, workflow design, and systems thinking applies directly.

I don't just design what users see. I think about how information moves through a system, how decisions cascade, and how the structure should behave when the edge case arrives.

System components

Input structure and classification User intent detection Dynamic prompt scaffolding Response evaluation layer Assumption extraction Tradeoff framing logic Action plan generation Session state and artifacts Feedback loops

What this connects to

This is the same systems thinking that I applied in healthcare data work — mapping information flows, identifying where data breaks down, designing around edge cases, and building structures that support real behavior instead of assumed behavior.

The difference between a well-designed AI experience and a bad one is often not the model. It's the architecture around it.

What a well-designed AI decision experience should make possible.

AI Compass is a product concept, not a shipped product. The outcomes below describe what the design was built to enable — what should get better when the system works as intended.

Reduced blank prompt anxiety

Structured intake replaces the intimidating empty field with a guided conversation that anyone can start.

Better decision framing

Multiple pathways prevent the first-answer trap — users compare options instead of defaulting to the obvious path.

Increased output trust

Assumption visibility lets users validate before acting — and gives them something to push back against when the AI is wrong.

Faster path to action

Action plan generation at the end of each session closes the loop between response and next step.

Reusable thinking

Session artifacts accumulate into a personal decision archive — the product gets more valuable with use.

What I would validate first: Does guided intake measurably reduce blank prompt anxiety compared to an open chat interface? This is the most testable claim in the product and the one that most directly proves the core design thesis.

What designing AI Compass taught me about what AI UX should actually become.

AI Compass challenged me to design past the chat interface — the thing most AI products are built around — and think about what it would mean to design the system that supports thinking, not just the widget that generates output. That's a more interesting design problem, and a harder one.

The project connected directly to my background in systems thinking, data workflows, and product design. The invisible infrastructure — intent classification, prompt scaffolding, assumption extraction, session artifacts — is what makes the visible interface trustworthy. You can't design the surface without understanding the system underneath it.

Adaptive intake depth: The intake process should be shorter for experienced users and more thorough for first-time or high-stakes decisions. Designing for this difference requires a better model of user context than a single flow provides.
Cross-session learning: If the product accumulates session artifacts, it should get better at surfacing relevant prior decisions when a similar problem appears. That's a hard information architecture problem worth solving.
Domain-specific loops: A standup retrospective decision loop is structurally different from a career pivot decision loop. Domain-specific templates would reduce intake friction for common decision types without sacrificing the framework.