Life Navigator

Inspiration

When a serious life event happens, you usually feel clueless and miserable in front of the bureaucracy machine. Suddenly you need to understand forms, deadlines, agencies, rights, and rules which nobody explained before. You do not ask "which government service do I need?" You ask: "what the hell should I do next?" Life Navigator was built to help people understand the situation, make a plan, and regain some control.

What it does

Life Navigator starts from a normal human sentence and turns it into an explainable roadmap.

The MVP supports three scenarios:

• Expecting a child in Israel
• Losing a job in Israel
• Relocating from Israel to the United States

It asks only questions that can change the plan, then shows what to check next, why it matters, when it may matter, and where to verify it. It is not another chatbot which gives a long answer and disappears. The roadmap is the actual product.

How I built it

The app is a Next.js and TypeScript application with three main building blocks:

• Event packs: reviewed definitions for every scenario, including questions, sources, tasks, rules, timing, and safety boundaries
• Deterministic roadmap compiler: turns validated answers into a changing plan
• Bounded AI layer: helps understand the initial user message and guide the conversation

The important part is that the model cannot invent official tasks, sources, timing, or eligibility results. Validated event packs and the compiler make those decisions. This is less magical, maybe, but also less likely to send someone to fight a ministry because an AI felt creative.

Challenges I ran into

The main challenge was not building a chat interface. It was keeping the product useful without pretending it can give legal, medical, tax, immigration, or benefits decisions. Another challenge was one-week hackathon logic: build something that works now, but do not hardcode one beautiful demo which dies immediately after the judges leave. So I focused on three real scenarios using one reusable engine, with source-backed content, deterministic rules, and clear verification boundaries.

Accomplishments that I am proud of

I had never used Codex before this project. The goal was to build something meaningful from zero code, using Codex and GPT-5.6 for discovery, planning, implementation, testing, refactoring, and deployment. I gave the direction, reviewed the work, and approved the important decisions. Codex did not become the CTO. Not yet.

I also tested Codex features as part of the real workflow:

• Repository-local skills for spec-driven development and life-event authoring
• Git worktrees for isolated research, content review, and deck preparation
• Remote control from my phone to follow the main Codex session, review work, give context, and redirect it away from disasters

The repository includes a dated CODEX_COLLABORATION.md record with the work, my decisions, changed artifacts, commits, and test evidence.

What I learned

My main learning was that Codex can support almost the full software-development lifecycle, not only generate code when you already know exactly what to build. I started with problem discovery and specs, then used Codex with GPT-5.6 for architecture, planning, implementation, testing, refactoring, deployment, documentation, and demo preparation. I wrote repository skills to make this less random: one skill defined spec-driven development, and another defined how to research, review, and author new life-event packs.

What's next for Life Navigator

I plan to continue this project after the hackathon. Next steps are adding more scenarios, better source-review workflows, more jurisdictions, and more personalized planning. Later, with proper permissions, security, expert review, and real integrations, Life Navigator could help users perform actions too. Not just show which blanks they need to fill. Because people should not need to become bureaucracy experts just to continue with their lives.

Built With

• codex
• gpt-5.6