Grounded
Questions stay tied to the designer’s project context.
2026 / DESKTOP APP · 0→1 · STRATEGIC DESIGN · PROMPT ENGINEERING
Rehearsl
Practice your pitch with voice based AI-powered stakeholder simulations.
ROLE
Product Designer, Human-AI Interaction
DURATION
5 days, Independent Project
RESPONSIBILITIES
Product Strategy, Competitive Research, Human-AI Workflow Design, Interaction & Visual Design, Prompt Engineering, Prototyping, AI Behavior Testing & Iteration.
Project Overview
Rehearsl is an AI-assisted coaching tool that helps product designers practice stakeholder presentations before the real meeting. Designers provide their project context and the stakeholders they expect to present to, then rehearse with role-aware AI personas that adapt their questions based on each stakeholder’s priorities and what the designer says.
The most important question in building an AI tool isn't what the AI does. It's what the human never gives up.
Context
THE PROBLEM
The design decision stays the same. The rationale needs to translate.
Designers can do strong work and still struggle to communicate it to stakeholders who aren’t designers. Explaining design decisions often means translating design rationale into language that connects with different priorities, yet there is little opportunity to practice those conversations before the actual meeting.
EVIDENCE
This communication challenge extends beyond individual presentations. In a 2026 Nielsen Norman Group analysis of about 150 responses from designers and researchers, roughly half pointed to alignment as their biggest challenge on product teams. Participants described difficulties influencing decisions, translating design work into stakeholder language, and navigating cross-functional priorities.
Research takeaway:
The challenge was not design craft alone. Designers also needed stronger ways to communicate, influence, and build alignment across disciplines.
THE OPPORTUNITY
Existing tools support presentation delivery, slide creation, or general communication skills, but they rarely recreate the specific conversation a designer is preparing for.
The opportunity was to create a rehearsal experience grounded in the designer’s own project and the stakeholders in the room, where the conversation can adapt as the designer responds.
Final Experience
The final Rehearsl experience brings the Human-AI framework into one end-to-end rehearsal flow. Designers provide the context of their project and the stakeholders they are preparing for, confirm the AI’s understanding, rehearse an adaptive stakeholder conversation, and review feedback grounded in what they actually said.
PREPARE
The designer provides the project context, who will be in the room, and what each stakeholder cares about. Rehearsl supplements this with relevant public company and role information where appropriate.
CONFIRM
Before the rehearsal begins, the designer reviews and corrects Rehearsl’s understanding of the project and stakeholders. I added this confirmation step so the AI does not begin coaching from unsupported assumptions.
REHEARSE
Questions are not predetermined. At each turn, the AI uses the stakeholder’s priorities, project context, conversation history, and the designer’s latest response to determine what to ask next.
REVIEW & FEEDBACK
Feedback is tied to what the designer actually said rather than generic presentation advice, helping them see where their reasoning was clear and where stakeholder priorities may have been missed.
OUTCOME
Stakeholder personas ask questions one at a time rather than following a fixed script. For each turn, the AI weighs the stakeholder’s priorities, the project context, and what has already been said to generate the most relevant next question. Because the model is probabilistic, the conversation can develop differently depending on how the designer responds.
Human-AI Framework
DEFINING HUMAN AND AI OWNERSHIP
I mapped the rehearsal process by separating what the designer knows from what the AI can reasonably infer.
Designer Owns
Stakeholder context
Stakeholder priorities
Interpersonal dynamics
Final judgement
AI Augmentation
Project synthesis
Public company/role context
Stakeholder simulation
Adaptive questioning
Response evaluation
The AI supports the rehearsal, but the designer keeps control of the context and judgment.
DEFINING AI BEHAVIOR
Before writing prompts, I defined what the AI needed to do for the rehearsal to feel relevant and realistic. A useful simulation should ask questions grounded in the designer’s project, reflect the priorities of each stakeholder role, and adapt to what the designer says. A weak simulation would rely on generic questions or invent assumptions about stakeholders that the designer never provided.
These expectations became the behavioral specification for Rehearsl: what the AI needs to know, how it should respond, and what it should never do.
Testing & Iteration
Because Rehearsl depends on both interface decisions and AI behavior, I evaluated more than whether the flow was usable. I also looked for moments where the AI made unsupported assumptions, behaved inconsistently with the intended experience, or gave the designer too much certainty about what it knew.
TESTING THE AI LOGIC BEFORE BUILDING
Before finalizing the prototype, I reviewed the system behavior against the rules I had defined earlier.
This surfaced three issues: the AI implied it could remember information beyond the current experience, inferred stakeholder dynamics the designer never provided, and introduced scoring before the evaluation criteria were clear.
I revised the logic so Rehearsl made fewer assumptions, avoided promises it could not reliably keep, and grounded feedback in clearer criteria.
TESTING THE PROTOTYPE
Once the full rehearsal flow was functional, new issues appeared in context.
The AI began coaching while it was still supposed to be confirming context. Speech recognition could distort important project details. An early version also previewed upcoming stakeholder questions before the designer responded.
Those findings changed the experience: I tightened the boundary between setup and coaching, made misunderstood context easier to correct, and removed question previews so the rehearsal stayed unpredictable.
KEY TAKEAWAY
A believable AI experience depends on more than generating good responses. It also needs clear rules around when the AI should act, what information it can trust, and what it should reveal to the user.
When the AI is Wrong
Rehearsl is a probabilistic system: the same pitch can get a different score and different feedback on a different run, because the personas generate their responses rather than following fixed rules. That variability is the product's reality, so the design treats every output as something to inspect, not a verdict to obey.
Rehearsl role-plays stakeholder personas and scores your pitch into structured feedback. That feedback can miss. A persona can judge your pitch on something it invented, give you a low score with a generic reason, or push you to satisfy one stakeholder in a way that would hurt the actual design. If the tool only said "trust the score," it would train designers to please an AI instead of thinking clearly.
I built three safeguards into the feedback:
Every score shows its reasoning, so you can see what the persona reacted to and judge whether the feedback is fair.
You can push back. If a critique does not apply, respond and make your case, and the practice continues from your answer instead of resetting.
The feedback is a training partner, not a judge. You stay in control of what to accept or reject, and the goal is a stronger second attempt.
Reflection
Rehearsl changed how I think about designing with AI. The challenge was not simply generating better responses, but deciding what the AI should know, what it should never assume, and where the designer needs to stay in control.
Building it also surfaced three tensions I would continue thinking about as the product evolves:
1. Useful context vs. sensitive context
More context can make the rehearsal more relevant, but it can also encourage designers to share sensitive project or organizational information. The challenge is finding enough context to be useful without requiring more disclosure than necessary.
2. Rehearsal vs. prediction
A realistic simulation can help designers prepare, but it should not imply that AI can predict how a real stakeholder will respond. The goal is to build confidence without creating false certainty.
3. AI availability vs. human mentorship
AI makes rehearsal available on demand, but practicing with colleagues also builds relationships and transfers organizational knowledge. Rehearsl should support that preparation, not quietly replace it.
Next Steps
1. Validate repeated use
So far, I have only tested Rehearsl on my own pitch practice. The next step is testing it across multiple rehearsals and real stakeholder meetings, to see whether practice improves how designers communicate decisions over time.
2. Test across contexts
Evaluate whether the feedback remains useful across different projects, stakeholder roles, and organizational settings.
3. Refine trust and personalization
Improve how Rehearsl handles sensitive organizational context, model uncertainty, and long-term personalization without creating false confidence about what the AI knows.
VIEW OTHER WORK
