About WhatWay

WhatWay answers questions about clinics, schools and legal or shelter services using a list of real places that a person has checked. The software never writes an address or a phone number: those come straight from the record.

Last reviewed 12 October 2026.

What we work on

Finding help in the United States means knowing which organisation does what, whether it will see you without insurance or papers, and whether anyone there speaks your language. That information exists, but it is scattered across agencies, out of date as often as not, and written for people who already know how the system works.

Asking a general-purpose chatbot is worse than it looks. It will answer confidently with a clinic that closed, or a phone number it assembled from nothing. For someone deciding whether to spend a bus fare they do not have, a confident wrong answer costs more than no answer.

So WhatWay is built the other way round. We keep a directory of verified services, and the assistant may only answer from it. It matches your question against that list, and when nothing matches well it says so and points you to 211 rather than offering you its best guess.

How a listing gets in

Every change is approved by a person. Nothing is auto-published.

1. Sources. HRSA health-center data, government and nonprofit listings, and forms submitted by clinics themselves.
2. Agents ingest. Entries are standardised, and duplicates and out-of-area listings are flagged.
3. Automated checks. Name, ZIP scope, phone number and website are validated.
4. Human review. Every change is approved by a person before it goes anywhere.
5. Published. Approved changes are deployed to the directory and live on the next query. No retraining, no waiting.

Clinics and users can report a wrong listing from any result on the site. Every report goes to a person for review.

Where we cover

The verified source datasets held 2,566 listings across five states as of 12 October 2026, with duplicates removed.

California

1,197 listings

New York

467 listings

Illinois

454 listings, 72% of them in Chicago-metro ZIP codes

Indiana

244 listings

Colorado

204 listings

By service, that is 2,479 health care, 39 education, and 48 resettlement, legal and shelter. The imbalance is real and worth saying plainly: health care coverage is deep, and the other two categories are thin. We are working on them.

The number on the home page is what you can search right now, which is whatever has been deployed. Reviewed source data can run ahead of the live site between releases, so the two will not always agree.

How it is built

Your question goes to a retriever that combines keyword and semantic matching, ranked by ZIP proximity, service, language and hours. The matching listings are then described in plain language in your language.

The language model runs on our own server rather than a commercial API, so queries stay in house and there is no per-query fee. It writes the sentence around the facts. It does not write the facts. That separation is the whole design: the model cannot invent a clinic because it is never asked to produce one.

This site is the public way in. A WhatsApp version runs against the same search pipeline and opens when Meta's verification clears. The interface is available in 28 languages, and cities are mapped to the states they are in, so searching by the name you actually use finds the same places as searching by ZIP.

Who builds it

Mowaffak Alraiyes

Northwestern University, Evanston, IL

Eric Jia, MS

Northwestern University Feinberg School of Medicine, Chicago, IL

David Liebovitz, MD

Northwestern University Feinberg School of Medicine, and the Institute for Artificial Intelligence in Medicine (I.AIM), Center for Medical Education in Data Science & Digital Health, Chicago, IL

WhatWay is incubating at The Garage at Northwestern. Nonprofit status filing and Meta business verification are both in progress, so we are not a registered 501(c)(3) yet and the public WhatsApp number is not open.

Before any formal enrollment, staff at RefugeeOne tested the prototype and reviewed the evaluation materials. Their feedback changed how resource navigation works, how directly you can get to a location, and how the survey questions are worded.

The study

WhatWay is being evaluated, not just shipped. The pilot has a Northwestern IRB Exempt determination and will run with 15 to 30 participants: adults 18 or over, English-speaking, who identify as refugees, immigrants, asylum seekers, or members of underserved communities.

Participants complete a survey after using it, covering usability, relevance, trust, confidence, privacy, and whether they would use it again, with room for open-ended feedback. The point is to find the answers people misread or do not trust, not to collect a score.

On privacy: prompts and responses are logged anonymously for the pilot study only. The public service does not store message text or phone numbers. The privacy policy covers what the live site does in full.

What we do not know yet

An honest list is more useful than a confident one:

• No outcome data. We have not yet measured whether using WhatWay means someone actually reaches care. Feasibility and trust come first; outcomes are a later study.
• The evaluation starts in English. The interface is in 28 languages, but the first round of testing is with English-speaking adults, which is a real limit on what it tells us.
• Listings go stale. Hours change and programmes close between verification cycles. Always call before you travel.
• The pilot is Chicagoland. The directory covers five states; the evaluation does not.

Tell us when we are wrong

Every result on the site has a link to report a problem with it, and that report goes to a person. You can also email whatway2026@gmail.com.

If the site is not working for you at all, call 211. It is free, answered by a person, and has interpreters.