Case Study 05

Building a decision-making tool with AI

Designed and prototyped Five Smart Friends — a tool that gives you five expert perspectives on any life decision.

Role: Designer & Developer

Type: Personal Project

Status: In Progress

Five Smart Friends landing page — hero section with bold typography and grid-paper texture

Situation

Rethinking how humans interact with AI beyond chatbots

I've been interested in how people make good decisions for a long time, and I kept hitting the same wall with AI: one voice, one perspective, one thread of advice. That's not how good decisions get made.

The insight was that we now have humanity's greatest thinkers at our fingertips — a Marcus Aurelius, a Ray Dalio, an Esther Perel can essentially walk into your living room and weigh in. Not just rounded perspectives, but incredibly good ones. Five Smart Friends gives you five expert lenses on any decision, each seeing something the others miss.

Action

Five distinct perspectives, not five variations of one

The core design challenge was differentiation. If you ask five AI personas the same question, they tend to converge on similar advice. The tool would only be valuable if each friend genuinely saw different things — the Philosopher notices what the Economist misses, the Creative reframes what the Relationship Expert takes at face value.

I designed five friends, each with a specific analytical lens, a distinct voice, and explicit instructions to disagree with the others when warranted:

The five friends — Philosopher, Economist, Relationship Expert, Creative, and Strategist with icons and descriptions

The five perspectives: Philosopher (meaning & values), Economist (costs & trade-offs), Relationship Expert (people & ripple effects), Creative (reframes & possibilities), Strategist (action & path forward).

Architecture

Designing the conversation, not just the interface

The product's quality lives or dies in the prompt architecture. Each friend has a system prompt that defines their analytical framework, their voice, and crucially — what they're supposed to notice that the others won't. The Economist is explicitly told to surface opportunity costs and second-order effects. The Relationship Expert is told to map human ripple effects the user hasn't considered.

The five prompts fire in parallel against the Claude API, and the Strategist's prompt includes a synthesis step that weighs the other four perspectives into a ranked recommendation. The user gets five distinct takes plus a concrete action plan — in under 60 seconds.

Interactive Demo

Decision Making Tool

TRY IT YOURSELF

Pick an example scenario and watch five friends weigh in. Each brings a different lens to the same situation.

5 perspectives + synthesis

Analyzing your situation...

Results

A working prototype with a clear product direction

5

Distinct AI perspectives
per query

<60s

Full analysis
with all five friends

4/5

Users found at least one
new angle they'd missed

The landing page and prompt architecture are complete. The next phase is building the full input-to-response flow as a working web app (Next.js + Claude API) and deploying it as a publicly accessible prototype.

This project demonstrates end-to-end product thinking — from identifying a real human need, through visual design, prompt engineering, and technical architecture, to a testable prototype.

Next Project

Building visual tools to improve vibe coding →

Amazon · Internal Tool