Skip to content
Syntasa

State Health & Human Services

A front door to care, in the words people use.

A Gemini-powered agent that helps people find local treatment in their own words and language, anonymously, from a database your agency vets. In production for the State of Arizona.

The problem

A complex system, when people are most open to help.

Millions of Americans live with opioid use disorder, and opioid overdose has been a declared public health emergency since 2017. Finding the right local care is part of the problem.

  • Word of mouth and outdated guides

    People seeking treatment relied on referrals, out-of-date community resource guides, or the work of navigating complex health plan websites.

  • No single guide for providers

    Within the health care system, providers often struggled to give patients appropriate resources, with no single-source guide to the options.

  • Language, insurance and stigma

    The people who need help most are often the ones facing barriers of language and insurance, at the moment they are open to asking.

  • Little visibility for the agency

    Information was not centralized and data analysis was limited, which made it hard to see treatment gaps and plan strategically.

How it works

From a question to the right door.

One pipeline from a person's question to an answer they can act on, and from their usage to the agency's planning.

  1. Step 1: Ask in your own words

    Anonymously, by text or on a phone: "I need help with opioid addiction," or "Where can I find methadone treatment?"

  2. Step 2: Understood in any language

    Google Translate lets the agent understand and instantly translate queries in multiple languages.

  3. Step 3: Matched to vetted providers

    Gemini on Vertex AI reads the question and answers from a database the agency has vetted, never from the open web.

  4. Step 4: Directions on any device

    Google Maps pinpoints nearby options and gives directions on a phone, tablet or computer.

  5. Step 5: Gaps become visible

    Looker dashboards show what people ask for and where, so the agency can see usage trends and gaps in service.

What it does

Built for the people who need it most.

  • Plain-language questions

    People describe their situation the way they would say it, and the agent interprets it, for example: "I'm pregnant and live in Apache Junction, and I need treatment and a place to live."

  • Many languages

    Integrated with Google Translate, so a person is met in the words they use.

  • Only providers in good standing

    The database lists only programs in good standing with the agency, which removes the worry about fraudulent or ineffective services.

  • Refine by what matters

    Narrow results for pregnancy, single parents with young children, or residential facilities that accommodate pets, a factor that matters for people experiencing homelessness.

  • Payment options

    Goes beyond location to show how care can be paid for, across publicly funded programs and private health plans.

  • Any phone, any hour

    Optimized for smartphone and SMS, working 24/7 for urban and rural users alike. Over a third of visits start on mobile.

Proof

In production for the State of Arizona since late 2021.

  • Unique individuals20,000+reached, expanding access to treatment information
  • Page views100,000+on the Opioid Service Provider Locator since launch
  • Engaged sessions55%+of sessions engaged: users find it valuable and use it
  • Cities120+across the state of Arizona

“Gen AI at the front door allows us to respond to the person looking for treatment resources in ways that meet them where they are, with what they know, in words they use, in ways that sound like and feel like help–and get accurate information to them in that moment of time when they are looking.”

Kate Dobler, State Opioid Treatment Authority, Arizona Health Care Cost Containment System (AHCCCS)

“You can engage anonymously and what comes back is a conversational set of information that's helpful, responsive, supportive, and not stigmatized. It's a front door into this ideally much larger array of information that anybody can interact with.”

Kate Dobler, AHCCCS

AHCCCS is Arizona's Medicaid agency, serving more than 2 million Arizonans. Referrals come from over 37 healthcare agencies and 80+ websites. Figures and quotes from Google Cloud's published AHCCCS story and video.

Read the case study on Google Cloud →

Built on

Google Cloud, end to end.

Arizona built it with Google Cloud and Syntasa. Firebase and Firestore keep the information current and scalable, and BigQuery holds the database of providers and resources that Vertex AI and Gemini answer from.

Products used in Arizona

  • Vertex AI
  • Gemini
  • Google Translate
  • Google Maps
  • BigQuery
  • Firebase
  • Firestore
  • Cloud Run
  • Google Analytics
  • Looker Studio

Trust

Information people can rely on.

A tool for finding care has to be trustworthy about what it is and what it isn't.

  • Information, not advice. The agent presents available options. It does not endorse a provider or a course of treatment, and it gives information, not medical advice or treatment recommendations.
  • Vetted listings. Only programs in good standing with the agency are listed, and providers keep their own listings current.
  • Anonymous by design. People can ask without identifying themselves, and the answers are conversational, supportive and free of stigma.
Security & Trust

The live locator states it plainly: the service is made available to help the public locate general information about treatment and services, and including a provider is not an endorsement of its quality or competence. It does not provide individual advice or counseling.

From the AHCCCS service locator

FAQ

Questions agencies ask first.

Bring a front door to care to your state.

Tell us the services people struggle to find and what your agency already vets. We will walk through what a locator agent would look like for you.

Talk with our teamSee the Arizona storyA working session with our engineers, on your data and constraints.