Align with Ash

Complete

Product Design

Prompt & Conversation Design

Development

Role/Contract

Product Designer

Developer

Team

Josh Hsu

Jason Liggi

Caitlin Stamatis

Neil Parikh

Timeline

Spring 2026

Tools

Claude Code, PostHog, Loops

Overview

What if getting two people in the same room at the same time for couples-therapy simply wasn't a requirement?

What if getting two people in the same room at the same time for couples-therapy simply wasn't a requirement?

Ash is an AI for mental health and personal growth built for individuals. Align with Ash asked whether that same product could be expanded to hold space for two at the same time.

Product Strategy

A standalone experience mirroring couples-therapy in real life, without the hassles of traditional therapy.

A standalone experience mirroring couples-therapy in real life, without the hassles of traditional therapy.

Backed by Clinical Psychologists

In collaboration with clinical psychologists, we introduced new couples-therapy modalities to our proprietary AI model.

Fast Feedback Loops

Built and maintained on the web, feedback & modifications could be implemented in the matter of minutes.

Problem

Couples wait an average of 6 years after problems start before they ask anyone for help.

Around 65 million couples in the US are in distress; only 20–30% ever seek help.

High Cost

$150–300 a session, rarely insured, with no guaranteed outcome. Financial stress is itself a top-five relationship problem

Lack of Access

Late entry is the single biggest predictor of treatment failure, and the long waitlists for traditional therapy don't help.

Partner Compliance

If one partner won't go, it defeats the purpose of couples therapy.

Problem

Couples wait an average of 6 years after problems start before they ask anyone for help.

Around 65 million couples in the US are in distress; only 20–30% ever seek help.

High Cost

$150–300 a session, rarely insured, with no guaranteed outcome. Financial stress is itself a top-five relationship problem

Lack of Access

Late entry is the single biggest predictor of treatment failure, and the long waitlists for traditional therapy don't help.

Partner Compliance

If one partner won't go, it defeats the purpose of couples therapy.

Problem

Couples wait an average of 6 years after problems start before they ask anyone for help.

Around 65 million couples in the US are in distress; only 20–30% ever seek help.

High Cost

$150–300 a session, rarely insured, with no guaranteed outcome. Financial stress is itself a top-five relationship problem

Lack of Access

Late entry is the single biggest predictor of treatment failure, and the long waitlists for traditional therapy don't help.

Partner Compliance

If one partner won't go, it defeats the purpose of couples therapy.

Opportunity

The bar to clear isn't a great therapist. It's Reddit at 2am.

The bar to clear isn't a great therapist. It's Reddit at 2am.

People who can't access therapy are already getting relationship help from forums, short-form video, self-help books and friends. Those sources are free, immediate, and frequently terrible. There's a real need for a product/experience that follows a clinical framework without the prerequisites.

Solution

Align with Ash

Align with Ash

Two private rooms, a common space, and your therapist's notes.

demo

Align with Ash (intro)

No app download required

No app download required

Unbiased moderation based on clinical psychology frameworks

Unbiased moderation based on clinical psychology frameworks

Asynchronous across time-zones and devices

Asynchronous across time-zones and devices

Projected Metrics

340 Couples Mediated

Per 1000 sessions that were started.

65% App Conversion Rate

Observed click-through from the session summary, a free session becomes the top of the subscription funnel

2.2x Confidence Resolving Conflict

Couples rated their ability to resolve conflicts before and after a single session from 14% to 31%

Process

We built the first prototype around one couple, then framed our asks.

We built the first prototype around one couple, then framed our asks.

In order to understand what our product constraints were, we had to build a narrative around a typical couple. We could categorize our asks into 4 groups.

Clinical Framework

Designing an unbiased mediator means choosing a school of therapy that works across a range of couples.

Designing an unbiased mediator means choosing a school of therapy that works across a range of couples.

Initially, I chose EFT as the core therapeutic modality because its clear stages are easy to turn into instructions.

But, pure EFT kept asking people to go deeper instead of naming what it saw, and sessions went in circles.

So, IBCT, Gottman and SFBT were blended in for directness, recognizable pattern language and brevity.

Select a modality to demo

EFT

IBCT

Gottman

SFBT

Prompt Architecture

The mediator's behavior is itself a product surface, and it facilitates productive conversation.

The mediator's behavior is itself a product surface, and it facilitates productive conversation.

How Ash behaves; whether it quotes you, who speaks next, how it ends, lives in the prompt. So I engineered it as a chain of five prompts, each with one job, passing structured information along.

Constraints

With safety & privacy at the forefront of this experience, how can AI bring together two individuals' contexts?

With safety & privacy at the forefront of this experience, how can AI bring together two individuals' contexts?

Safety

Couples therapy isn't recommended when there's an abusive partner.

We engineered the prompt to provide an emergency route and crisis resources upon detecting signs of abuse.

We engineered the prompt to provide an emergency route and crisis resources upon detecting signs of abuse.

Privacy

Protecting users' privacy every step of the way.

Ash lets each user choose what they don’t want to share with their partner. Private details stay in their private room. Anything shared in the commons stays there. After each session, memory is erased within 24 hours.

Legal Framing

Ash is an experimental wellness tool, not a licensed therapist.

We went screen by screen with legal counsel and added friction (user consent required to proceed) to screens that carried legal weight.

Product Architecture

Getting two busy people to participate in therapy at once is a huge source of friction that needs to be addressed.

Getting two busy people to participate in therapy at once is a huge source of friction that needs to be addressed.

We tested a few interaction models to decide the final form of this experience.

Why the web?

The key decision was not building it in our app.

The key decision was not building it in our app.

Reluctant partners won't install an app, but will click direct links

This also allows for users with different devices to participate.

No scheduling required

As each web session remains active for 24hrs, partners can come and go as they please, across time-zones.

Web sessions do not have access to anyone's mental health history

This allows for much more unbiased representation coming in and leaving.

No time lost to release-cycles

Prompt changes, UI fixes and A/B tests can shipped immediately without app-store reviews.

Final Flow

1 — Invite your partner

2 — Talk to Ash privately

3 — Ash moderates in the commons

4 — Ash's notes for the couple

UX Decisions

In a room, you can tell when the therapist is about to speak and who they're talking to. How do you do it online?

In a room, you can tell when the therapist is about to speak and who they're talking to. How do you do it online?

I added typing states, a waiting screen before The Commons, Ash's opening message broken into short beats, and a color for each speaker.

Should users be allowed to interrupt?

Should users be allowed to interrupt?

The hand-raise button was distracting when always visible, and gave people nowhere to hold what they wanted to say. I fixed it to appear only after Ash responds, with space to draft a reply.

What was left out

During tense moments, should couples off load their conversation?

During tense moments, should couples off load their conversation?

Forced timeouts were rejected by our clinical advisor, since a pause mid-argument feels like being cut off.

Should we show Ash's reasoning ?

Should we show Ash's reasoning ?

A panel showing Ash's reasoning would risk couples arguing with the mediator instead of resolving the issue at hand.

Launch

Proving it was safe, then releasing it to the world.

Proving it was safe, then releasing it to the world.

Conducting over 10 jailbreak scenarios, aimed at the privacy promise

Every weakness we found along the way became a guardrail in the prompt.

A custom Analytics pipeline for a product with two users per session

Built in PostHog, without needing cookies or accounts, keeping survey answers anonymous and any email address stored separately.

Emailed to Ash users who had opted into experimental features

Using Loops, we emailed a select group of Ash users; who had opted in to try experimental features, expecting rough edges and, willing to give feedback.

Organic traction on LinkedIn without paid promotion

A single post from our team about Align reached 13,700 impressions on a network of about 2,500 connections. Soon our inbox was flocking with invite rquests.

Reflection

During a company offsite, Josh Hsu & I were talking about a marketing video idea; showing how the typical Ash-user might complain about their partners to Ash.

Why can't Ash just directly talk to a user's partner to solve their issues?

Why can't Ash just directly talk to a user's partner to solve their issues?

Josh being a lawyer, brought his unique perspective on how mediation between 2 parties works in real life. Caitlin helped us pick and chose what therapeutic modalities make most sense. Jason helped me set up Ash locally so I could work on the prompt and run the product by myself.

Rethinking Distribution

Gathering 2 individuals rather than 1 is not a piece of cake!

Bearing in mind that this product is meant for two individuals, we would have to come up with a marketing strategy that caters to couples primarily.

Bearing in mind that this product is meant for two individuals, we would have to come up with a marketing strategy that caters to couples primarily.

Prompt Engineering

Shaping this version of Ash

This was the first time I had the opportunity to change our model's prompt hands-on. I spent a week making Ash more articulate, when the real fix was making it quieter and handing the work to the couple.

This was the first time I had the opportunity to change our model's prompt hands-on. I spent a week making Ash more articulate, when the real fix was making it quieter and handing the work to the couple.

Privacy & Legal

Can you trust AI to be 100% private & secure?

It's nerve-wracking to know that individuals might share intimate details with Ash that they wish to keep completely private from their partner. We did our best to set up guardrails for possible scenarios, but there may still be edge cases that break the threshold.

It's nerve-wracking to know that individuals might share intimate details with Ash that they wish to keep completely private from their partner. We did our best to set up guardrails for possible scenarios, but there may still be edge cases that break the threshold.

Overview

Problem

Opportunity

Solution

Impact

Process

Clinical framework

Prompt Architecture

Product Architecture

Final Flow

UX Decisions

Launch

Reflection

AKR

Design

Play

About

Contact

AKR

Design

Play

About

Contact

Design

Play

About

Contact