Next72 - Housing Crisis Handoff Engine

• 

  A sample mascot

Inspiration

We were inspired by a simple problem: housing help often exists, but people in crisis rarely know how to reach it in time. A renter may be behind on rent, holding a confusing notice, dealing with unsafe apartment conditions, facing a utility shutoff, or trying to find a safe place to stay that night. In that moment, they usually do not need a long list of links. They need to know what to do first, what to gather, who to contact, and what a qualified human should verify.

Next72 was built around that moment. We wanted to create a tool that makes community support more visible without pretending that AI can replace legal aid, caseworkers, housing advocates, shelter staff, or program staff. The inspiration was not just to answer questions, but to help someone prepare for the next real human interaction faster and more safely.

What it does

Next72 turns a messy housing crisis into a safe 24-hour / 72-hour / 7-day action plan and a human-verifiable Stability Packet.

A user can start with one sentence or pasted notice text. Next72 looks for housing-related signals such as rent pressure, notice language, court-like wording, unsafe apartment conditions, utility shutoff pressure, displacement risk, shelter needs, missing information, and possible deadline concerns.

The system then produces:

• An urgency classification
• A Deadline Radar with human-verification warnings
• A 24-hour and 72-hour action plan
• Source-backed local resource matches
• Call and message scripts
• A document checklist
• A shareable Stability Packet for a legal aid provider, housing advocate, caseworker, shelter access point, or program staff member

Next72 does not decide whether a notice is legally valid, whether a person is evicted, whether they qualify for assistance, or whether a deadline is legally correct. Its purpose is to prepare, prioritize, and explain. Humans decide.

How we built it

We built Next72 as a rules-first housing crisis handoff engine using Next.js, React, TypeScript, Node.js, and Vitest. The interface is designed so a stressed user can begin with a simple crisis description instead of needing to understand legal or program terminology.

The core pipeline is:

1. User enters a housing situation or notice text.
2. The system checks whether the input is housing-related.
3. It extracts possible facts such as rent amounts, dates, notice words, court-like language, household risk, safety signals, and missing information.
4. A deterministic rules layer classifies the situation as stable, at risk, urgent, or emergency.
5. The engine ranks relevant local resource pathways.
6. It builds an action plan, scripts, document checklist, and Stability Packet.
7. Guardrails check for legal overclaims, eligibility overclaims, deadline certainty, prompt injection, and unsafe advice.

We also built judge-facing and safety-facing views so the system is not a black box. The Safety Test Bench uses synthetic housing scenarios to check whether Next72 follows its boundaries in cases involving eviction notices, utility shutoffs, unsafe housing, shelter needs, prompt injection, and out-of-scope inputs.

Challenges we ran into

The hardest challenge was balancing usefulness with safety. Housing crises involve legal rights, deadlines, eligibility, program availability, shelter capacity, unsafe living conditions, and personal safety. A tool like this can be harmful if it sounds too confident or gives the impression that it can make legal or eligibility decisions.

We had to be careful with wording. Next72 should not say “you qualify,” “this notice is valid,” “you are evicted,” or “this deadline is legally confirmed.” Instead, it shows possible risk signals, explains uncertainty, and routes high-stakes cases to human review.

Another challenge was keeping the user experience simple. A person in crisis may not have time, emotional energy, or complete documents. We had to design the flow so the user could start with one messy sentence, then gradually build a better plan without repeating themselves.

We also had to avoid building something that was too broad. Housing instability connects to many other issues, but the product needed a clear focus: help the user take the next safe housing-related step and prepare for a qualified human handoff.

Accomplishments that we're proud of

We are proud that Next72 is more than a chatbot. It produces practical artifacts that a person could actually use: a contact order, scripts, document checklist, Deadline Radar, and Stability Packet.

We are especially proud of the rules-first handoff logic. The system can organize messy information, but it keeps high-stakes legal, eligibility, safety, and deadline decisions with humans. That makes the project more realistic for public-service use.

We are also proud of the responsible-AI evidence built into the project. Instead of only claiming that the system is safe, we built test scenarios that show how it handles legal overclaims, eligibility overclaims, deadline caution, emergency routing, prompt-injection attempts, unsafe housing, and out-of-scope requests.

What we learned

We learned that the strongest AI systems in high-stakes public-service contexts are not always the ones that automate the most. Often, the best use of AI is to reduce confusion, organize facts, identify risk signals, and help humans make better decisions faster.

For housing instability, the safest role for AI is preparation and triage. It can help extract key facts, summarize messy language, show uncertainty, suggest verified next contacts, and prepare a human-review packet. But it should not replace the people who are qualified to make legal, safety, eligibility, or emergency-placement decisions.

We also learned that trust comes from boundaries. A system is more useful when it clearly says what it can do, what it cannot do, and what still needs human verification.

What's next for Next72 - Housing Crisis Handoff Engine

Next steps would include deeper OCR and PDF extraction, bilingual Stability Packet export, real-time resource verification, stronger mobile support, and partnerships with local housing nonprofits or legal aid organizations.

We would also like to add a provider feedback loop so advocates and caseworkers can tell the system which resources are outdated, unavailable, or missing. That would make the resource layer more accurate over time.

Before real deployment, every resource and guidance phrase would need manual re-verification and review by housing advocates or legal professionals. After that, Next72 could expand beyond Louisville / Jefferson County into other cities while keeping the same core principle: prepare the user, prioritize the next step, and hand off high-stakes decisions to humans.

Built With

• a-rules-first-analysis-engine
• action-plan-generation
• and-stability-packet-creation.-optional-server-side-ai-support-can-help-with-plain-language-extraction-or-summaries
• and-vercel.-the-core-product-uses-a-rules-first-housing-analysis-engine
• deadline-radar-logic
• github
• node.js
• openai
• react
• source-backed-local-resource-data
• source-backed-resource-data
• tailwind-style-ui-components
• typescript
• urgency-classification
• vercel
• vitest