Inprove
Text-based, AI-powered, mental-health support
inprove is an AI-powered self-improvement platform that guides users through their individual problems, utilizing personalized exercises and recommendations pulled from a library of the most effective mental health resources found online.
try inprove
Role: Co-Founder, Head of Design
Website: inproveapp.com
“It's not therapy, but it's built with real guidance from therapists.
Inprove is an AI assistant you can message whenever you need support to reset, think clearly, and act differently. No apps, no logins, just simple text-based support designed to help you actually shift what’s keeping you stuck.”
Text-based theraputic guidance facilitated by LLM, pulling from a database of the top exercises, resources, and guidance recommended by professionals online. Users test out short, focused exercises drawn from evidence-based therapies like CBT, ACT, and Internal Family Systems, and our system creates actional recommendations and reminders to help people change.
Delivered on-demand via WhatsApp, Telegram, or SMS – all for free.
Objective
Build a mental health product for people who don’t have access to therapy.
We came into this with a fairly simple goal, can we make something effective and affordable? Each of us had seen the benefits that come from therapy but are also very aware of the lack of availability.
Everyone seems to have a reason not to go to therapy, but everyone also seems to know how helpful therapy is...
Our hope is to fix that.
Product Evolution
Resource Sharing (User Generated Content)
(January 2024)Our first web app was a text-heavy platform, similar to Twitter, where users and professionals could post their favorite exercises for different mental health topics. The structure was an open blog format where users could describe the exercises and write about their experiences. Navigation around the application was primarily organized by the mental health area they were interested in learning more about.
The hope was that users and professionals would share their top exercises, giving students free access to high-quality tools and exercises. The benefit for professionals would be a funnel for business. If students found someone they resonated with, it might eventually lead them to book with that therapist.
what we learned
-
Wrong pain point, saving needed to be streamlined and simplified for people to actually participate
- Lack of resonance with professionals despite the common desire for more resources and a funnel for new business
- Professionals utilize their close network for discovery much more than any other channel
-
More interest in top resources provided by professionals, ie. top books, podcasts, articles, etc.
version 2
Public Access Database + Quick Uploads
(April 2024)We built off the foundation of our last product for our second release, but aimed to simplify it, focusing on two main areas: helping users save and organize resources, and making it easier to discover useful ones. In interviews, we consistently heard that people were saving social media posts, articles, or videos, but they could never find them again when it mattered. We wanted to streamline this by offering a fast and simple way to save and organize a link.
On the discovery side, we saw that users trusted their friends and community far more than strangers or random professionals. Our application counted how many times a unique resource was saved to act as a signal of popularity and usefulness. We also introduced a lightweight social layer, where users could “friend” others and see a feed filtered by what their network found helpful. Additionally, they could search by topic or resource type when looking for something specific.
This was our first product launched on the Apple App Store and Google Play Store, but was unavailable on the web.
what we learned
- Context matters, it can’t just be the resource by itself, each person’s story and background matter
- Scrolling fatigue, people have no interest in endlessly scrolling another platform
- Need for specificity, we can’t offer topics so broad that people can’t relate to them directly
- Reluctance for people to upload resources themselves, but loved saving what other people saved
INTERMISSION
Key insight: Storytelling
An important pivot came from a series of workshops where we spent time analyzing different aspects of storytelling and narrative building – because each therapy session is after all just one long story being told. One narrative structure I particularly enjoyed was Aristotle’s 7 Elements of Good Storytelling, pictured below. It helped provide not only a structure but essentially a check list for the things we needed to bring to each interaction on our platform.
It got interesting when we started mapping those narrative elements to the elements we offered on our platform, and helped us to identify which parts felt strong and where we needed to add. One thing we realized we were missing was the emotional aspect. A lot of this content is helpful, but it’s being passively consumed, and a lot of the emotion is lost in that kind of interaction–It’s something you get every time from a therapist.
From there I thought it would be helpful to map out what some of these conversations look like.
i.e. if I was meeting a new therapist, what would I say to them? What would I need to tell them in order for them to be able to appropriately help me?
And if they asked follow up questions about one topic in specific, how much additional information would they need to know? How do I get them caught up as quickly as possible? Turns out it doesn’t take too much information, and when it’s broken down it maps to our database categories pretty easily.
what we learned
- narrative drives so much more than it appears in theraputic settings
- someone’s personal narrative contains a huge amount of easy sorting data
- utilizing other’s narratives could provide the most helpful recommendations
END INTERMISSION
version 2.1
Database Search + Active Problem Solving Using Reviews
(June 2024)Our second iteration then expanded, we launched a more robust platform that was accessible across desktop, iOS, and Android. This version was built around an open-ended search experience, users could now start by asking an actual natural language question like “What can we help you with?” or “What do you want to improve?” Instead of navigating by category alone, we invited people to describe their own problems or goals, then we would match them to resources in our database, or automatically pull new ones from across the internet to fill out our database.
We layered in a light recommendation system, pulling from how often something had been saved or engaged with across the platform. On top of that, we started experimenting with review aggregation, pulling in feedback from across the internet to give users a better sense of which tools were most trusted, and making direct recommendations for users.
This version marked a shift: we weren’t just organizing tools anymore — we were trying to help users solve their problem directly.
what we learned
- Context matters, it can’t just be the resource by itself, their story and background matter
- Scrolling fatigue, people have no interest in endlessly scrolling another platform
- Need for specificity, we can’t offer topics so broad that people can’t relate to them directly
- Reluctance to save resources themselves, but loved what other people saved
IMPORTANT
Right around this time was the introduction of the OpenAi api and the ability to create custom agentic systems. However, bad press about chatbots making easily avoidable mistakes means people are still very skeptical of using them.
3rd iteration
Chatbot + Custom Modules
(November 2024)First and foremost, this version of the product was our first time incorporating an LLM, leaning on it to deliver the emotional aspect we were missing in earlier versions. It worked like a standard chat, but leaned on our database of resources and reviews to satisfy the user’s query or help generate solutions to their problems, utilizing custom designed modules that could be generated on the fly by the agent.
This phase was less of a clean product release and more of a working concept that kept evolving as we dug into what we were actually trying to build. The goal was to deliver insights dynamically, based on where someone was in their own process. We created a framework based around the time that someone could dedicate to the platform, to make sure we were helping people find resources that matched their current needs, but didn’t lock them into something for longer than they had hoped. There were a lot of questions around how to define search parameters, interpret user input, and generate helpful outputs in real time.
Resource saving was still part of the experience, but we layered on context: what symptom or challenge is someone facing? What kinds of tools have helped others in similar situations? We began experimenting with early versions of personalized recommendations based on a user’s history, goals, and content that others found useful via online reviews.
what we learned
- Much more fluid delivery of insights and seemed to actually help people
- Still a lot of user reluctance to using chatbots, society was not on board yet
- Overall most testers were drawn to this interaction vs the manual search platform
4th iteration
Dynamic Content Delivery + LLM Integration
(February 2025)
Our next release brought everything together into a more guided and personal experience, all within a mobile app. This version was structured around three core elements: topic, insight, and experiment. In this hybrid experience users would start by chatting with the app in a familiar chat-style interface, where they could talk through what was going on, reflect on recent challenges, or explore areas they wanted to work on. From there, the chat would offer direction: suggestions for experiments, tools to try, or new perspectives to consider.
Once a decision was made, users would leave the chat to enter a separate focused interactive experience where they could apply what they were learning. Kind of like a minigame. After finishing the experiment, they'd return to the chat to debrief and reflect, building a record of personal insights over time. We also introduced a new dashboard to help users track their progress, revisit past moments, and start to see patterns in their journey.
This version was mobile-only and initially deployed through Expo Go as we tested and de-bugged, but we were really trying to decide whether or not to launch. Something felt off and we finally realized the hard truth that no one wants or needs another app to download – it’s a lot of friction for something that needs to be easy.
what we learned
- No one wants to download another app (unfortunately)
- Content that is located and delivered by an agent feels comfortable for users
- Users reported real life progress when interacting with the agent, rather than passively consuming
- Public opinion of chatbots is quickly improving, not weird for people anymore at all
TODAY
Direct Messaging
(June 2025)Our final (and current) version of the app made one major shift, no more app. After testing and working with users, we realized that the most helpful element of our entire platform was the conversation and practical experimentation, none of which actually requires our own custom interface. This version of the product runs entirely on existing platforms, right now being WhatsApp, Telegram, and via SMS texts – with plans (if there’s market fit) to also run on platforms like Slack or Teams or Discord.
The core structure of the product is the same: identify the user’s problem, provide related resources and information, guide them through an exercise, analyze the effectiveness, and either find something that works better, or save it for later so you’ve always got something available to help you calm down.
It’s honestly a pretty cool product, and it’s all entirely free – as tools like this should be.
inproveapp.com
Outcome
3 product releases
100+ hours of user interviews
Acquiring new users / active marketing + advertising
Collaboration
Balancing individual visions for the product, realistic business and development goals, and user experience expectations.
Collaborated, designed, and architected the platform data systems, working to always find the most efficient solution to whatever problem was at hand.
Managed + collaborated with external contractors on the first product build and supervised additional contractors for smaller, late-stage projects.