Case Study
0
0
•
Designer & Builder
•
Self-Initiated
•
1 week
•
2026
Global Trend Engine
A self-initiated futurology intelligence dashboard powered by AI agents that scan the web for frontier signals, synthesize cross-domain patterns, and generate predictive convergence insights. Built entirely from scratch — concept, design system, React architecture, prompt engineering, API integration, and deployment.
The Challenge
Tracking meaningful signals across seven global domains — AI, climate, biotech, geopolitics, energy, society, and space — requires reading dozens of sources constantly. There was no single tool that gathered these signals, synthesized patterns across them, and pointed toward probable future developments. Existing news aggregators surface what's happening now — but they don't synthesize signals across domains or point toward what might be coming next.
The Solution
An AI-powered dashboard where seven named domain personas (Turing, Gaia, Mendel, Caesar, Faraday, Orwell, Sagan) scan the live web for frontier signals, a synthesis layer identifies cross-domain patterns, and a predictive agent called Nostradamus generates a convergence insight with an estimated timeframe — visualized as a convergence timeline diagram.
"The gap between 'designer who uses AI tools' and 'designer who builds AI products' is narrower than most people assume — and the design skills most relevant to agentic systems are the ones we already have."
— Daniel Rivas, case study reflection
3 Agents
Linear pipeline: scan → synthesize → predict
~45s
Full 7-domain scan
2 Deployments
Targets (Claude + Vercel)
The Process
01
Discovery & Research
Identified a gap in available tools for tracking meaningful signals across multiple global domains simultaneously. Existing aggregators surface current news but don't synthesize across domains or predict future convergence points.
Needs
A single interface that gathers frontier signals from live web sources, synthesizes them across domains, and surfaces predictive patterns — without requiring hours of manual research.
Frontier signal: A real, current development (research finding, policy shift, product launch, or emerging phenomenon) that indicates meaningful change in a domain's trajectory.
Findings
No existing tool combined multi-domain signal gathering with cross-domain synthesis and predictive insight.
AI agents with live web search could replace hours of manual monitoring.
Naming agents after historical figures (Turing, Gaia, Mendel) created an intuitive mental model for me (the user).
02
Definition
Designed a system of three agents in a linear pipeline: a scanning agent with seven named domain personas, a synthesis agent (Nostradamus), and a meta-prediction agent (Tarot). The seven personas are structural roles within the scanner, not independent processes — a deliberate UX decision to create an intuitive mental model.
Opportunity
Claude's web search capability and API made it possible to build a fully agentic scanning system that retrieved real, current frontier signals rather than relying on training data.
Personas
Self — as a futurology enthusiast tracking global trends across AI, climate, biotech, geopolitics, energy, society, and space.
Pain Points
The practice of tracking frontier signals always occurs in hindsight.
There's no single place to gather insights from various segments or extrapolate predictions based on that data.
03
Design & Prototyping
Built the full interface in React with the portfolio design system applied throughout. Designed three real agents in a linear pipeline, seven named domain personas, a domain toggle system, model selector, signal cards, synthesis panel, and the Nostradamus convergence timeline visualization.
Ideation
Built the Agent System
Horizon runs on three agents in a linear pipeline. Seven named personas — Turing, Gaia, Mendel, Caesar, Faraday, Orwell, and Sagan — are structural roles within a single scanning agent, not independent processes.
Turing — AI & Computation
Gaia — Climate & Earth Systems
Mendel — Biotechnology & Health
Caesar — Geopolitics & Power
Faraday — Energy & Infrastructure
Orwell — Society & Culture
Sagan — Space & Frontier
After the initial scanning design, I added a second distinct agent to synthesize the data into a predictive outcome based on future trends.
Nostradamus — Predictive Synthesis (2nd iteration, separate API call)
After building prediction history, I added a third distinct agent to read across accumulated predictions and generate a higher-order meta-insight.
Tarot — Meta-Prediction (3rd iteration, separate API call)
Convergence Timeline
An SVG diagram using straight right-angle elbow routing — horizontal lines turning 90° at junction points into a vertical spine, then a trunk arrow into the Nostradamus rounded-rect node — communicating how independent domain signals converge toward a single predicted future state.
Prototyping
In this case study, design and prototyping worked as one, prototyping designs in realtime, testing them, and iterating. This has become one of the new adaptations of using AI tools in my design work.
04
Delivery
Deployed as a React artifact inside Claude (live API calls, no external hosting required) and as a standalone Vercel app with a serverless API proxy that keeps the Anthropic API key server-side.
Result
A fully functional futurology intelligence dashboard with live web search, cross-domain synthesis, and predictive convergence visualization. Runs in approximately 30–60 seconds for a full 7-domain scan.
Ultimately, I chose to support only the Claude native version as a personal tool instead of support a publicly accessible deployment through Vercel — Claude APIs can be expensive to support for a public deployment with no monetization built-in.
I also iterated on the design to:
Automate a nightly scan
Store the history of predictions, exportable to an Excel file
Use a 9th meta-prediction agent (Tarot) to read the history of Nostradamus summaries and create a new prediction from those. Tarot identifies recurring themes, escalating patterns, and higher-order convergences.
Lessons learned
Prompt engineering is interface design — the quality of every output is determined by how the model is instructed.
Agentic UX requires designing for uncertainty — every state (idle, scanning, cancelled, errored, complete) needs deliberate design.
Naming creates meaning — agent names with historical resonance give the system character and help users build mental models instantly.
Visualization is synthesis, not decoration — the convergence timeline communicates the Nostradamus function more clearly than any prose description.
Prototyping happens in realtime as part of design iteration in the new world of AI-native design process.
Selected Screens & Artifacts
The system overview
Zero state, ready to run based on the selected model
Sample of the Nostradamus summary, front and center
Tools Used
Daniel Rivas · UX Strategy & Product Design
← Back to home
The Challenge
Tracking meaningful signals across seven global domains — AI, climate, biotech, geopolitics, energy, society, and space — requires reading dozens of sources constantly. There was no single tool that gathered these signals, synthesized patterns across them, and pointed toward probable future developments. Existing news aggregators surface what's happening now — but they don't synthesize signals across domains or point toward what might be coming next.
The Solution
An AI-powered dashboard where seven named domain personas (Turing, Gaia, Mendel, Caesar, Faraday, Orwell, Sagan) scan the live web for frontier signals, a synthesis layer identifies cross-domain patterns, and a predictive agent called Nostradamus generates a convergence insight with an estimated timeframe — visualized as a convergence timeline diagram.
"The gap between 'designer who uses AI tools' and 'designer who builds AI products' is narrower than most people assume — and the design skills most relevant to agentic systems are the ones we already have."
— Daniel Rivas, case study reflection
Outcomes
3 Agents
Linear pipeline: scan → synthesize → predict
~45s
Full 7-domain scan
2 Deployments
Targets (Claude + Vercel)
The process
1
Discovery & Research - Summary
Identified a gap in available tools for tracking meaningful signals across multiple global domains simultaneously. Existing aggregators surface current news but don't synthesize across domains or predict future convergence points.
Users
Self — as a futurology enthusiast tracking global trends across AI, climate, biotech, geopolitics, energy, society, and space.
Needs
A single interface that gathers frontier signals from live web sources, synthesizes them across domains, and surfaces predictive patterns — without requiring hours of manual research.
Findings
No existing tool combined multi-domain signal gathering with cross-domain synthesis and predictive insight.
AI agents with live web search could replace hours of manual monitoring.
Naming agents after historical figures (Turing, Gaia, Mendel) created an intuitive mental model for users.
2
Definition - Summary
Designed a system of eight AI agents — seven domain specialists and one predictive synthesis agent — each with a named identity tied to a historical figure relevant to their domain expertise.
Opportunity
Claude's web search capability and API made it possible to build a fully agentic scanning system that retrieved real, current frontier signals rather than relying on training data.
Key Design Decisions
Apply existing portfolio design system (DM Serif Display, DM Sans, #F8F7F4, #1a56e8) for visual consistency.
Expose model toggle (Sonnet vs Opus) as a first-class UI element — treating cost and quality as design dimensions.
Signal intensity bars in sidebar show relative domain momentum, not just binary signal presence.
Convergence timeline uses circuit-board elbow routing to communicate how signals route into a single prediction.
3
Design - Summary
Built the full interface in React with the portfolio design system applied throughout. Designed three real agents in a linear pipeline, seven named domain personas, a domain toggle system, model selector, signal cards, synthesis panel, and the Nostradamus convergence timeline visualization.
Agent Architecture
Horizon runs on three distinct agents, not eight. Seven named personas operate as structured roles within a single scanning agent — they share one model context and one API call. The distinction is intentional: this architecture delivers most of the value of a true multi-agent system at a fraction of the cost and complexity.
Scanning agent — one API call, seven domain personas (Turing, Gaia, Mendel, Caesar, Faraday, Orwell, Sagan)
Nostradamus — separate API call; synthesizes all domain signals into a predictive convergence insight
Tarot — separate API call; reads accumulated prediction history to generate a higher-order meta-prediction
Personas vs. Agents
Naming seven domain roles as distinct agents (Turing, Gaia, etc.) is a deliberate UX decision, not a literal description of the architecture. The names create an intuitive mental model and make the system feel intentional — but Horizon is honest about what it is: three real agents in a linear pipeline, with seven personas providing structure inside the first.
Convergence Timeline
An SVG diagram using straight right-angle elbow routing — horizontal lines turning 90° at junction points into a vertical spine, then a trunk arrow into the Nostradamus rounded-rect node — communicating how independent domain signals converge toward a single predicted future state.
4
Delivery - Summary
Deployed as a React artifact inside Claude (live API calls, no external hosting required) and as a standalone Vercel app with a serverless API proxy that keeps the Anthropic API key server-side.
Result
A fully functional futurology intelligence dashboard with live web search, cross-domain synthesis, and predictive convergence visualization. Runs in approximately 30–60 seconds for a full 7-domain scan.
Lessons Learned
Prompt engineering is interface design — the quality of every output is determined by how the model is instructed.
Agentic UX requires designing for uncertainty — every state (idle, scanning, cancelled, errored, complete) needs deliberate design.
Naming creates meaning — agent names with historical resonance give the system character and help users build mental models instantly.
Visualization is synthesis, not decoration — the convergence timeline communicates the Nostradamus function more clearly than any prose description.
