Landed: moving to Abu Dhabi, in the right order

Hub71+ AI Hackathon · team landed

The problem. Everything a newcomer needs already exists: ACCESSADGM, ICP, TAMM, UAE PASS, SEHA, banks, schools, ILOE and Hub71. It is scattered, so people find each tool too late or in the wrong order. A spouse's visa waits on the employee's Emirates ID, which waits on the medical test, which waits on the company's establishment card.

What Landed does. It builds one personal plan from a rules graph that links each person, their family and their company. Each step opens the existing tool at the right moment. Landed brings no new government system.

What runs in this demo

• Rules engine (DAG). Finishing a step recalculates what unlocks for you and for the people linked to you, such as your spouse or your company. Every unlock has a reason and a source.
• Personal agent on OpenAI (/api/ask).
  • The agent knows this person's live plan: stage, what is done, what they can do now, who they are waiting on.
  • It answers in any language and suggests one next action for that person.
  • Facts come only from verified cards with a source. If no card answers, it refuses instead of inventing.
  • If the API is unreachable, an offline matcher answers instead.
• PII firewall before the LLM. The pipeline is Presidio-style: analyze → anonymize → LLM → deanonymize.
  • Names (deny-list recognizer), emails, phone numbers, Emirates ID numbers (784-…), IBANs, passports and cards become tags like <PERSON_1> before OpenAI sees the text.
  • Real values are restored only in the reply to the user.
  • The UI shows exactly what the model saw.
  • POC, not for prod: production runs Microsoft Presidio (analyzer + anonymizer, spaCy NER) as a Docker sidecar.
• Districts. Areas are ranked against the office and your priorities. Labels carry a source and follow the law: no protected attributes, and member labels only appear at k≥10.
• Company view, operator dashboard (anonymous funnels) and safety layer.
• Life after 90 days. Covers changing jobs, the grace period, ILOE, renewing a visa and leaving the UAE.

Data

55 person steps, 7 company steps and 16 verified answer cards, each with a source URL and a verification date. The data is hand-curated test data for the demo.

Run / deploy

• Static index.html plus one Vercel serverless function, api/ask.js.
• Set OPENAI_API_KEY (and optionally OPENAI_MODEL, default gpt-4.1-mini) in Vercel → Settings → Environment Variables. See .env.example.
• No secrets live in the repo.