Inspiration
26 million Americans are credit invisible because they don't exist in the financial system. 80% of 18-year-olds have no credit history at all. The disparity is stark: 15-16% of Black and Hispanic consumers are credit invisible versus 9% of white consumers. In low-income neighborhoods, that number hits 30%.
This isn't just a credit problem — it's an information access crisis. 59% of college students have considered dropping out due to financial stress. $46 billion in scholarships goes unclaimed every year. Only 21% of first-gen students receive help completing FAFSA. And only 27% of Gen Z can define the word "deductible." The financial system was built for people who already know how it works and it punishes everyone else.
We know because we lived it. We're both first-gen, underserved students at UMich. We lost scholarship money we didn't know we qualified for. We signed leases with clauses that turned out to be illegal under Michigan law. We went years being credit invisible without knowing it was a fixable problem. The information existed but it just never reached us.
That gap between "the information exists" and "the people who need it have it" is what Navi closes. Not with lessons. Not with another budgeting spreadsheet. With AI agents that connect the dots no one connected for us such as rent reporting to credit building to mortgage eligibility to down payment assistance. They act on your behalf so you don't have to figure out a broken system alone.
What it does
Navi is an AI-powered financial life platform that maps every financial decision a Gen Z user faces into a single interactive graph and deploys AI walker agents to find the optimal path to any goal.
Core Platform: Financial Life Graph — Your credit score, student loans, rent, savings, scholarships, government benefits, insurance, and homeownership goals are all nodes in one connected graph. Edges represent how decisions compound: paying rent on time → credit score increase → mortgage eligibility → down payment assistance qualification. PathfinderWalker — An AI agent that traverses your personal graph to find the fastest sequence of actions to reach any financial goal, generating personalized advice at every node using your actual numbers. EligibilityWalker — Scans 10+ real programs (MSHDA DPA, Pell Grant, SNAP, Medicaid, LIHEAP, AOTC, Esusu, Work-Study) and matches you based on your profile. Maya, our demo user, has $3,000+/year in unclaimed benefits. SimulationWalker — "What if I save $200 more per month?" Instantly projects your financial future and shows the timeline shift: mortgage ready 20 months earlier, $24,000 less rent paid. HealthScoreWalker — Computes a 0-100 financial health score across savings rate, credit, debt-to-income, rent burden, and emergency fund status. Gamification — XP system, achievement badges, streaks, and level progression that makes financial progress feel like a game, not a chore. AI Agents That ACT:
AI Lease Analyzer — Paste any lease, get instant clause-by-clause analysis flagged red/yellow/green with specific state statutes cited. Generates dispute letters for illegal clauses. In our demo, a sample Ann Arbor lease has 4 illegal clauses including a $150/day late fee (Michigan caps at ~$25-50) and a clause restricting 911 access. Application Auto-Filler — One click fills an entire rental application from your verified profile. No more re-entering the same data across 15 apartment applications. Scholarship Auto-Applier — Matches you to scholarships by eligibility score, drafts personalized essays using your profile, and queues applications. Our demo found 4 scholarships totaling $12,000 for Maya's profile. Insurance Gap Analyzer — Only 21% of Gen Z has renters insurance, and only 27% can define "deductible." This agent identifies coverage gaps and explains each insurance type in plain English. Navi Chat — Always-on conversational AI advisor that references your specific numbers. Ask "How do I build credit?" and get an answer based on YOUR income, YOUR rent, YOUR situation and not generic advice.
How we built it
Backend: Jac (Object-Spatial Programming) The entire intelligence layer is built in Jac. We chose Jac because a financial life is inherently a graph problem: decisions are nodes, consequences are edges, and finding optimal paths requires traversal — exactly what walkers do. Our navi.jac file contains: 7 node types: UserProfile, FinancialGoal, ActionStep, Resource, Milestone, InsuranceNeed, LeaseClause 6 edge types: HasGoal, RequiresAction, NextStep, LeadsToMilestone, HasMilestone, QualifiesFor 7 walker agents: PathfinderWalker, EligibilityWalker, SimulationWalker, HealthScoreWalker, InsuranceGapWalker, LeaseAnalyzerWalker, ChatWalker 15+ by llm() functions where the type signature IS the prompt — no manual prompt engineering needed Pre-seeded graph with 3 demo user profiles, 11 action steps across 5 financial paths, 10 real resources with real eligibility criteria, and 6 milestones The key architectural insight: the graph is the reasoning engine. When PathfinderWalker visits a LeaseClause node and traverses to jurisdiction-specific Statute nodes to identify violations, it's performing structured legal reasoning through graph traversal — not dumping text into an LLM and hoping for the best. This creates auditable, explainable AI decisions. Frontend: Vanilla HTML/CSS/JS The frontend is a single navi.html file, zero dependencies, zero build step, opens in any browser. Built with: Custom SVG graph visualization with animated walker traversal Real-time node interaction (click any node for AI-generated advice) 5 dashboard tabs: Path (graph), Agents (lease/app/scholarship), Resources (eligibility matches), Simulate (what-if projections), Badges (gamification) Split-panel layout with persistent AI chat sidebar Dark theme designed for Gen Z aesthetic
Challenges we ran into
Jac syntax evolution. Some syntax patterns from the documentation didn't work in v0.13.5.
LLM API access. We initially planned to use Anthropic's Claude API for all by llm() functions but needed to pivot to Groq's free tier (Llama 3.3 70B) when we couldn't get API keys quickly enough. The byllm package's model-agnostic design made this a one-line change — just swap the model name.
Accomplishments that we're proud of
The graph visualization is real. When PathfinderWalker animates across Maya's financial life graph, lighting up nodes and edges in sequence, it's not a mockup — it represents actual walker traversal logic from the Jac backend. Every resource is real. MSHDA Step Forward, Pell Grant, SNAP, Medicaid, LIHEAP, AOTC — these are actual programs with actual eligibility criteria. When Navi says "you qualify for $7,395/year in Pell Grants," that's based on real federal guidelines applied to the user's profile. The lease analyzer cites real Michigan statutes. MCL 554.633 (late fee limits), MCL 554.601 (entry notice requirements), MCL 600.2922 (constitutional rights) — these are real laws that protect real tenants in Ann Arbor. Zero dependencies on the frontend. One HTML file, no npm, no React, no build step. Double-click to open. The entire platform runs offline. The by llm() integration is clean. Function signatures like def generate_advice(user_profile: dict, action_step: dict, goal: dict) -> str by llm() let the type system do the prompting. No prompt templates, no string concatenation, no output parsing. It actually works. jac run navi.jac builds the graph, runs all 7 walkers, and produces real AI-generated output. The HTML frontend is a fully interactive product demo. Together, they tell a complete story.
What we learned
Jac's Object-Spatial Programming model is genuinely powerful for this problem domain. Financial life decisions are naturally a graph, not a table, not a document, not a conversation. Modeling them as nodes with typed edges and traversing them with stateful walkers produces reasoning that's more structured and auditable than raw LLM prompting. Gen Z's financial problems are interconnected, and no product treats them that way. Every existing tool solves one slice: Cleo does spending, Bilt does rent rewards, Scholly does scholarships. The data is devastating. 26 million credit-invisible Americans. $388 billion annual cost of financial illiteracy. 54% of student loan borrowers with anxiety directly from debt. 1 in 14 borrowers with suicidal ideation from financial stress.
What's next for Navi
Phase 1 — Product Completion (Next 3 months):
Open banking integration via Plaid for real-time financial data Live AI chat powered by Claude with full graph context Expand jurisdiction coverage for lease analyzer (all 50 states) FAFSA auto-fill agent with step-by-step guidance Rent negotiation agent with comparable market data Bill negotiation agent (internet, phone, insurance) "Financial Wrapped" — Spotify-style annual financial progress report
Phase 2 — University B2B Channel (Months 3-12): White-label version for university financial aid offices Integration with student information systems Retention analytics dashboard (59% of students consider dropping out from financial stress — Navi reduces this) Compliance with new financial literacy mandates (30 states now require financial education) Target: 10 university pilots, starting with UMich
Phase 3 — Scale (Year 2+): Employer financial wellness benefits (B2B PEPM model, $1.21B market by 2029) Credit union partnerships for customer acquisition Mortgage pre-qualification pipeline with referral revenue Investment readiness assessment and retirement starter Peer-to-peer financial mentorship marketplace International expansion (UK, Canada, Australia have similar Gen Z housing crises)
Log in or sign up for Devpost to join the conversation.