This README:

• Explains the philosophy
• Shows your architecture
• Mentions Kiro explicitly
• Matches the hackathon theme perfectly
• Reads like a real product

---

Next, we move to B — the full AWS Builder Center blog draft
I’ll generate it in a single, copy-paste-ready piece.

Reply:

RefereeAI – The Trade-off Referee

RefereeAI doesn’t tell you what to choose.
It shows you what you’re trading.

Most AI tools optimize for giving a single answer.
Real-world decisions—especially in engineering—are not about “best”.
They are about trade-offs: speed vs control, cost vs flexibility, simplicity vs power.

RefereeAI is a decision-support tool that compares options under constraints and explains what you gain and what you give up with each choice.

This project was built for Kiro Week 6 – “The Referee” challenge.

---

🧠 The Problem

Ask most AI tools:

“Should I use AWS Lambda or EC2?”

You’ll get a ranked list or a single recommendation.

But real decisions are not binary truths—they are value judgments:

• Do you value speed or control?
• Do you optimize for today or for scale?
• Are you a solo developer or an enterprise team?

AI should help you choose, not just consume answers.

---

💡 The Idea

RefereeAI treats AI as a neutral referee, not an oracle.

Instead of:

“Use Lambda.”

It says:

“If you choose Lambda, you gain speed and simplicity,
but you give up control and predictable latency.”

Every decision is framed as a trade.

---

🏗 Architecture

Browser UI
     ↓                        (constraints)
Decision Engine              (rules)
     ↓                        (structured trade-offs)
Kiro Prompt Compiler
     ↓
Narration Layer (Mock LLM / Kiro in production)
     ↓
Human explanation of trade-offs

Layers

1. Decision Engine (Deterministic)

   • Interprets user constraints
   • Computes gains & sacrifices for each option
   • Ensures decisions are structured and explainable

2. Prompt Compiler

   • Embeds engine output into a strict “Referee” prompt
   • Prevents the model from giving a single “best” answer

3. Narration Layer

   • In production: powered by Kiro / LLM
   • In this demo: deterministic mock for reproducibility
   • Converts structured trade-offs into human language

This separation ensures:

• Reasoning is testable
• Narration is pluggable
• The system never becomes a generic chatbot

---

🎯 What This Demo Does

RefereeAI compares:

AWS Lambda vs EC2
for a REST API backed by PostgreSQL

Based on:

• Traffic pattern
• Time to market
• Budget sensitivity
• Team size
• Architecture preference
• Risk tolerance

It returns:

• Trade-off summary
• What each option gains & sacrifices
• Persona-based perspective
• Regret preview

It never says “choose X”.

---

▶️ Run Locally

Requirements

• Node.js (LTS)

Steps

cd referee-ai/backend
node index.js


http://localhost:3000

Change constraints and click “Compare Trade-offs”.

You’ll see how different contexts produce different reasoning.




🧩 About Kiro

    RefereeAI is designed around Kiro’s strength in structured prompt orchestration.

    The .kiro/ directory contains:  

    A strict “Referee” prompt

    A workflow that enforces:

    No single answers

    Explicit trade-offs

    Persona-based framing

    Regret awareness

For hackathon reproducibility, this demo uses a deterministic narrator.
In production, this layer is replaced with Kiro / Bedrock / OpenAI to generate fresh explanations on every run—without changing the architecture.