SwarmLite

Project “Clean Room” · Built by Anthony Key

Test ideas against
synthetic market
intelligence.

Swarm Lite creates AI persona swarms that review your business proposals, debate your features, and forecast market temperature — before the LinkedIn post, before the reactive sprint, before you commit resources.

The Story

Why product teams keep reacting — and a better way forward.

5

AI personas per swarm

0

Manual research steps

100%

HITL approval on actions

Projects · markets · branches

The Problem

The reactive strategy trap

Product teams are reactive by default. A competitor ships a feature. Someone screenshots it into Slack. A meeting gets scheduled. A week later, a half-formed response lands in the backlog — with no user research, no strategic context, no assessment of whether responding is even the right move.

Monday

Competitor ships

A new feature appears on a competitor changelog. No one on your team knows yet.

Wednesday

LinkedIn screenshot

Someone posts about it. Your PM sees it and starts a Slack thread. Knee-jerk reactions begin.

Friday

Reactive sprint

A feature request lands in your backlog. No user research, no strategic analysis, no signal — just reaction.

Swarm Lite gives you the brief before the screenshot.

Market intelligence treated as a stateful system process — not a research exercise. Test your response before you build it.

Built for enterprise trust

Persona Swarms

Create AI personas representing target users and stakeholders. Run huddle sessions where they debate, critique, and stress-test your proposals.

Session Branching

Branch from any session to test variations — different personas, adjusted queries, follow-up questions. Preserve the original while exploring alternatives.

Market Monitoring

Define competitive landscapes with competitors, watch items, and scan intervals. The system monitors and surfaces relevant signals.

Artefact Reports

Every session produces a structured brief — overall sentiment, per-persona breakdown, key themes, and prioritised recommended actions.

Action Approval Gate

Resource-intensive decisions require explicit human approval. Every action logged with cost, status, and context. HITL by design.

Team Collaboration

Share projects, markets, and personas with specific team members or make them globally available across your organisation.

Thinking Behind the Build

Four decisions that shaped the architecture

01

AI personas over surveys

Surveys take weeks and suffer from response bias. AI personas give instant, structured feedback from validated archetypes — and you can branch and re-test in minutes, not months.

02

IBM Carbon over custom UI

Enterprise credibility requires enterprise design. Carbon provides the component system, spacing tokens, and accessibility compliance that make the app look like it belongs in a procurement conversation.

03

Branching over linear sessions

Real strategy is not linear. When a persona raises a security concern, you need to follow that thread with just the enterprise personas — without losing the original five-persona context.

04

Human-in-the-loop on all actions

AI proposes, humans decide. Every resource-intensive action surfaces with cost and context. The approval gate is not a limitation — it is the feature that makes this deployable in an enterprise.

Read the full architecture →

A chassis, not just a product tool

Anywhere you need structured feedback from representative user archetypes before committing resources — this architecture applies.

Product Strategy

Constraint: Feature prioritisation

Test proposed features against user personas before sprint planning

Marketing

Constraint: Campaign positioning

Validate messaging with target archetypes before launch

Healthcare

Constraint: Patient experience

Test service changes against patient persona panels

Financial Services

Constraint: Product suitability

Validate investment product positioning with client archetypes

Education

Constraint: Curriculum design

Test learning experiences against student and educator personas

Government

Constraint: Policy impact

Model citizen reactions to proposed policy changes

Modern, open, composable

Next.js 16 + React 19

Frontend

IBM Carbon

Design System

IBM Carbon

Design System

Python FastAPI

Backend

LangGraph

Agent Graph

Anthropic Claude

LLM

SQLite

Persistence

Carbon + Lucide

Icons

IBM Pictograms

Pictograms

View full architecture →

Part of a trilogy

The execution layer. The intelligence layer. The testing layer.

About the Builder

Anthony Key is a Lead UX Designer and Product Architect with a track record of translating complex operational problems into products people can actually use.

Swarm Lite is a proof of concept and a provocation: what if businesses could test their strategic decisions against a synthetic panel of representative users — instantly, repeatably, and with full transparency into how each persona reasoned?

Every architectural decision in this project is documented and defensible. This is not a side project. It is a portfolio artefact demonstrating AI product architecture, agentic systems design, and the kind of senior product thinking that connects technical choices to strategic outcomes.