CrossroadsAI
View project ↗An AI-powered decision council that dynamically assembles a personalized panel of expert agents (career advisors, immigration experts, financial analysts, psychologists) to help navigate major life choices through structured multi-perspective analysis.
Overview
CrossroadsAI is an AI-powered decision council that dynamically assembles a panel of specialized expert agents tailored to your specific situation. Rather than getting a single AI perspective on a complex life choice, you get a structured multi-perspective analysis from agents spanning the domains that actually matter for your decision.
Built in 1.5 hours on Manus, it is designed around a core insight: the most valuable thing about consulting real experts is not any single opinion, but the synthesis of competing considerations across domains. CrossroadsAI makes that structure available for any decision.
01. The Problem
Major life decisions almost always involve multiple competing dimensions that cannot be optimised independently. An international student deciding whether to stay in the US after graduation must simultaneously weigh career trajectory, visa status and immigration risk, financial runway, and psychological wellbeing. These factors interact in non-obvious ways.
Most AI assistants respond to complex decisions with a single synthesised answer, which collapses the tension between perspectives before you have a chance to think through each one. The result feels helpful but obscures the real tradeoffs.
02. How It Works
The user submits a decision as a plain-language query. The app reads the decision and dynamically assembles a council of specialist agents matched to the domain mix the decision requires. For an international student's career dilemma, the council might include a career advisor, an immigration expert, a financial analyst, and a psychologist.
Each agent analyses the decision independently, without seeing the other agents' responses, producing a perspective grounded in its own domain logic. A moderator agent then synthesises all perspectives into a structured recommendation that makes the tradeoffs and points of consensus explicit.
- →Dynamic council assembly: agent composition is inferred from the decision, not pre-set
- →Independent analysis: each expert reasons without anchoring to the others
- →Moderator synthesis: a final agent surfaces agreement, tension, and a structured recommendation
03. Architecture
This is a classic multi-agent orchestration pattern: a dispatcher layer reads the query and selects agent types, parallel agent invocations run independently, and a synthesis step aggregates outputs. The key product decision is keeping the synthesis as a separate agent pass rather than a post-processing step, which lets the moderator reason about disagreements rather than just average them.
The interesting design problem is not building the agents but deciding what the moderator should surface: where do the agents agree? Where do they conflict? What does the user most need to know to make the call?
04. Reflections
Building this in 1.5 hours required committing early to the dynamic assembly pattern instead of hard-coding agent types. That choice is what makes the product feel responsive to your specific situation rather than generic.
The design question I kept returning to: when experts genuinely disagree, should the moderator pick a side or surface the disagreement? CrossroadsAI surfaces the tension, which respects user autonomy but requires the user to do the final integration themselves. For high-stakes decisions, I think that is the right call.
