RCDRenel Chantes Domond

← Back to the blog

AI

How I built One Next Step

A reentry planning app, built with the same five steps I use with every business: gather, define, root cause, solve, own it.

By Renel Chantes Domond

One Next Step chat screen where the guide, Pip, asks what the person needs support with today
One Next Step starts with one question from its guide, Pip.

People coming home from incarceration don't lack resources. They lack a way to know which one comes first, and a way to see it through. In my AI Lab capstone, my teammates and I built a tool to fix that. It's called One Next Step.

I'm going to walk you through how we built it, using the same five steps I use with every business I consult for. Swap "reentry" for whatever problem keeps you up at night, and the process is the same.

  1. Gather

    Research, interviews, a caseworker and a survey.

  2. Define

    The whole problem in one sentence.

  3. Root cause

    Sequence and follow-through, not a lack of resources.

  4. Solve

    About ten iterations, then one clear next step at a time.

  5. Own it

    It runs on its own today.

1. Gather

I started where every good build starts: with the research. I spent the early weeks pulling the studies that mattered, and my team talked to the people who live this problem.

We ran three in-depth interviews with people who had come home, a questionnaire with a caseworker, and a needs survey of 16 people who were still incarcerated.

A few things came up again and again:

  • ID and documents come first. Without them, almost nothing else moves.
  • Transportation, phones and online applications stand in the way of everything else.
  • People leaned on family, not on formal systems.
  • None of the three people we interviewed had used 211, Connecticut's main resource line.
  • Everyone wanted a plan: a step-by-step list, not a pile of phone numbers.

The research backed it up. The first year home is the riskiest: of people released from state prisons in 2012, 37% were arrested in their first year, compared with 26% in their fifth. And of people released from federal prison in 2010, a third found no formal job at any point in the next four years.

2. Define

Our first draft of the problem was about jobs. The interviews pushed us wider. Jobs mattered, but so did housing, ID and health, and the order changed from person to person.

So we cut the problem down to one sentence:

People coming home have no coordinated way to identify their most urgent barrier, decide what to do first, and carry a referral through to a completed action.

That sentence did a lot of work. Every idea we had after that, we held up against it.

3. Root cause

Here's the part most people skip: we wrote down the evidence against our own idea. Employer discrimination is real. Housing waitlists are real. And people with a good caseworker often already have what they need.

What survived was sharper than where we started. The gap is biggest for people outside a case-managed program. And the failure isn't just knowing what's out there. It's sequence and follow-through. No birth certificate means no ID. No ID means no job. Miss the first step, and the whole chain stalls.

We were also honest about what a tool can't do. It can't end discrimination, shorten a waitlist, or replace a caseworker. It can help someone take the next right step, and then the one after that.

4. Solve

I was the build lead. I started by sketching what kind of tool might work, drawing on what I'd learned in years of advocacy work. Then I had Claude research options against the problem we'd defined. Once the team agreed on a direction, I built about ten iterations of the tool before the final build.

What we cut. Our early plan was text messaging, so it would work on any phone. In practice, it wasn't feasible in our timeline. So we switched to a web app that works in any phone's browser and can be saved to the home screen.

What we worked around. We wanted Connecticut 211's resource data, and approval was still pending. So the app pulls from Connecticut 211's public resource listings, which they keep current. The directory stays fresh without anyone updating it by hand.

What it does:

  • You say what you need help with, in your own words.
  • It builds a plan in the order that fits you, and you can move steps up or down.
  • It gives you one next step at a time: a real place to go and a real number to call.
  • If that option doesn't work out, it offers another one, not the same one again.
  • You check things off as you go: tried it, still waiting, got help.
  • Pip, a small shepherd companion, grows as you hit milestones like getting your ID or finding stable housing.

How the AI is used. The AI only sorts what you type into a type of need, and watches for a crisis. It doesn't pick providers or give advice. Simple fixed rules choose the resource. Emails, phone numbers and ID numbers are stripped out before the AI sees a message.

Safety and privacy came first. Crisis words bring up 911 and 988 right away. Typed messages aren't stored. Your plan is yours, and the app never asks for sensitive personal information.

5. Own it

Testers appreciated how simple it was. They could describe what they needed in plain words, get a plan built around their essential needs, and walk away with an actual next step and somewhere to turn.

Today, One Next Step runs on its own. Because it refreshes from 211's listings, it always has current places to send people. I step in when something needs changing. Otherwise, it just works.

Try One Next Step

What this means for your business

If you see your own business in this story, here's the lesson: don't start by building. Gather as much information as you can. Research it. Name the real problem. Then craft the simplest solution, build more than one version, plan for what could go wrong, and let real people test it before you take it to market.

That process turns a good idea into something proven to work. On a free call, we'll name the real problem in your business together.

Book a free 30-minute call

Questions, answered

Who is One Next Step for?

People coming home from incarceration in Fairfield County, Connecticut. Anyone can open it in a phone browser.

Does it replace a caseworker?

No. It helps people take the next right step, and it points everyone to real help.

What does the AI decide?

Only what type of need you're describing, and whether a message sounds like a crisis. Fixed rules pick the resource.

Built with my AI Lab teammates as our Cycle 4 capstone. See the One Next Step case study.

Sources: Bureau of Justice Statistics, Recidivism of Prisoners Released in 34 States in 2012 · Bureau of Justice Statistics, Employment of Persons Released from Federal Prison in 2010

This post is for educational purposes only. One Next Step is not monitored for emergencies; in a crisis, call 911 or 988.

Stay inspired. Get practical ideas.

Join the list for new insights and resources.