Arcel
Arcel
Helping hybrid athletes train with a coach grounded in 300+ sports-science papers.
Helping hybrid athletes train with a coach grounded in 300+ sports-science papers.
Timeline
Timeline
April 2026 – Present
April 2026 – Present
Team
Team
2 Engineers, 1 Designer
2 Engineers, 1 Designer
Role
Role
UI/UX Design & Product
UI/UX Design & Product
Skills
Skills
AI interaction design, Mobile UX, Information architecture, Prototyping
AI interaction design, Mobile UX, Information architecture, Prototyping
What is Arcel?
What is Arcel?
Arcel is a training app for hybrid athletes. It builds a plan around strength, conditioning and a sport, and answers training questions using 300+ sports-science papers.
Arcel is a training app for hybrid athletes. It builds a plan around strength, conditioning and a sport, and answers training questions using 300+ sports-science papers.
The session for the day
The session for the day
Arcel project requirement
Arcel project requirement
Working with two engineers, I was asked to turn Arcel from a research assistant into a training coach. This included a feature audit, a PRD, a new information architecture and a clickable prototype.
01
01
01
Context
Context
Context
Hybrid athletes balance strength, conditioning and a sport, so a plan has to work around games and practices.
Hybrid athletes balance strength, conditioning and a sport, so a plan has to work around games and practices.
02
02
02
Why it matters
Why it matters
Why it matters
Richard Xu and Dimitrios Mahairas had built retrieval over 300+ papers. It needed to feel like one coach, not five tools.
Richard Xu and Dimitrios Mahairas had built retrieval over 300+ papers. It needed to feel like one coach, not five tools.
03
03
03
Launch
Launch
Launch
Arcel is launching soon on the App Store.
Arcel is launching soon on the App Store.
The problem
The problem
Arcel began as a research assistant with a planner attached. Answers backed by real papers lived on their own page, so athletes had to move between tools to decide what to do each day.
The first version, called Shingo, shown with a sample question.
The first version, called Shingo, shown with a sample question.
How might we make sports-science research feel like a coach, not a library?
How might we make sports-science research feel like a coach, not a library?
How might we make sports-science research feel like a coach, not a library?
Users and stakeholders
Users and stakeholders
In this step, I looked at who Arcel is for and what they need from it.
In this step, I looked at who Arcel is for and what they need from it.
01
01
01
Know today’s session
Know today’s session
Know today’s session
One workout, fully prescribed.
One workout, fully prescribed.
02
02
02
Plan around games
Plan around games
Plan around games
Practices and matches shape the week.
Practices and matches shape the week.
03
03
03
Trust the advice
Trust the advice
Trust the advice
Every answer can show its sources.
Market research
Market research
I looked at six products people already use to train, and where each one leaves a gap.
I looked at six products people already use to train, and where each one leaves a gap.
Design process
Design process
I worked through questions like - When should the coach speak up? Who makes changes to the plan? How do people check the advice? The answers became three rules.
I worked through questions like - When should the coach speak up? Who makes changes to the plan? How do people check the advice? The answers became three rules.
01
01
01
Summoned, never ambient
Summoned, never ambient
Summoned, never ambient
Screens stay calm until you ask.
Screens stay calm until you ask.
02
02
02
No silent edits
No silent edits
No silent edits
Every plan change shows why it happened, with a before and after.
03
03
03
Evidence on demand
Evidence on demand
Evidence on demand
Any answer opens to the sources retrieved for it.
How the UI changed
How the UI changed
Arcel started as Shingo, a plain page for asking research questions. It became a dark web workspace, then a mobile app. The current version is a light app built around the day’s session, with chat as a button on every screen.
Arcel started as Shingo, a plain page for asking research questions. It became a dark web workspace, then a mobile app. The current version is a light app built around the day’s session, with chat as a button on every screen.
How Arcel pulls research
How Arcel pulls research
Arcel is only useful if people can trust where an answer came from, so the search behind it matters as much as the screens. Richard and Dimitrios built the system below. It has two parts: one builds the library ahead of time, and one runs every time someone asks a question.
Arcel is only useful if people can trust where an answer came from, so the search behind it matters as much as the screens. Richard and Dimitrios built the system below. It has two parts: one builds the library ahead of time, and one runs every time someone asks a question.
01
01
01
Why search twice
Why search twice
Why search twice
Meaning catches a question worded differently from the paper. Keywords catch exact terms, like an exercise or a protocol.
Meaning catches a question worded differently from the paper. Keywords catch exact terms, like an exercise or a protocol.
02
02
02
Why drop reference lists
Why drop reference lists
Why drop reference lists
A bibliography is full of matching words but holds no advice, so those passages are removed before they can be found.
A bibliography is full of matching words but holds no advice, so those passages are removed before they can be found.
03
03
03
Why an AI check
Why an AI check
Why an AI check
No single score means a result is good enough. The check reads what came back and decides, at the cost of one more model call.
No single score means a result is good enough. The check reads what came back and decides, at the cost of one more model call.
04
04
04
Why lock the model
Why lock the model
Why lock the model
The search model is pinned to one version. If the library was built with a different one, search stops instead of returning bad matches.
The search model is pinned to one version. If the library was built with a different one, search stops instead of returning bad matches.
Where the backend shaped the UX
Where the backend shaped the UX
How the system works decided how the screens behave.
How the system works decided how the screens behave.
01
01
01
Clean answers
Clean answers
Clean answers
The model leaves out citation markers, so answers stay easy to read on a phone.
The model leaves out citation markers, so answers stay easy to read on a phone.
02
02
02
Sources on demand
Sources on demand
Sources on demand
One button opens each excerpt with its page, license and link.
One button opens each excerpt with its page, license and link.
03
03
03
Quick or Deep
Quick or Deep
Quick or Deep
Quick uses 5 results and one retry. Deep research uses 10 and up to three.
Quick uses 5 results and one retry. Deep research uses 10 and up to three.
What it doesn’t do yet
What it doesn’t do yet
Sources belong to the whole answer, not to each sentence, and the list shows everything retrieved, not only what was used. Plans use research when they are made, but they don’t keep a record of it. Those are the biggest gaps for trust.
Sources belong to the whole answer, not to each sentence, and the list shows everything retrieved, not only what was used. Plans use research when they are made, but they don’t keep a record of it. Those are the biggest gaps for trust.
Features
Features
Onboarding answers, workouts and sports sessions are turned into a three-week plan that refreshes weekly.
Onboarding answers, workouts and sports sessions are turned into a three-week plan that refreshes weekly.
Plan the week
Plan the week
The user adds practices and games, and the plan works around them.
The user adds practices and games, and the plan works around them.
Ask the coach
Ask the coach
The user asks a question from any screen and can open the excerpts behind the answer.
Track progress
Track progress
The user sees twelve weeks of consistency and strength milestones.
The user sees twelve weeks of consistency and strength milestones.
Screens from the current app with sample training data. The chat answer is illustrative; the cited paper is real.
Screens from the current app with sample training data. The chat answer is illustrative; the cited paper is real.
Summary
Summary
Arcel went from a research assistant with a planner attached to a training coach built around each athlete’s week. I redesigned the information architecture and visual system, and the team built it into a working mobile app.
Arcel went from a research assistant with a planner attached to a training coach built around each athlete’s week. I redesigned the information architecture and visual system, and the team built it into a working mobile app.
What’s next
What’s next
We are still working on Arcel. Right now we are improving how plans are generated and how research is found and cited, and getting the app ready for the App Store.
We are still working on Arcel. Right now we are improving how plans are generated and how research is found and cited, and getting the app ready for the App Store.