Juno: the second brain that acts before you ask
Role
Founding Product Designer
Team
1 PM, 2 Engineers, 1 Designer
Timeline
Spring 2026, 10-week sprint
Skills
Product Design, Product Strategy, Branding
Overview
Founder first, designer second.
As the sole product designer in a team of one PM and two engineers, we were tasked with building something, from scratch, in 10 weeks. The definition of wearing many hats and doing more than just design.
The problem
Students have always scattered their own data, losing visibility of their own lives.
We talked to 50 students who shared a specific frustration: relying on Notion, G-Cal, Gmail (an assortment of productivity tools), to organize their lives, but find frustration in having to align themselves with many platforms before actually getting to work.
Students lose visibility of what’s truly important when their data is scattered.
Opportunity
We finally have the tools to give students that visibility back—without the manual work.
Visibility means something different for every student, whether it’s never missing a workshop or losing track of a deadline. Unlike tools that wait to be asked or require manual input, that visibility can exist by default and compile in the background.
AI finally reads and acts on context in real time.
Context
The longer you use conversational AI, the more it already knows and compute improves.
Knowledge Graph
Everything is added to one continuously growing map of your life, making traces significantly easier.
Solution
Juno: the second brain that acts before you have to ask...
A knowledge graph that compiles as you talk to it, reachable through a single text.
......quietly bringing visibility back to a life scattered across a dozen apps.
Core flows
Chat with real-time context
Juno responds with an understanding of your inbox, calendar, and anything you choose to connect. Receive briefs on what's relevant all through text, right at onboarding.
Look back on your room
Review your background automations in a place that feels like home.
Flip into your dashboard
In a visual dashboard, you can regain visibility on what's important to you, automated by Juno's context on you.
Research
Living inside the AI tools already claiming this space.
Granola, Notion AI, these tools were already used on the daily by our team. We dug specifically into Poke, a popular software amongst AI-native college students who applauded its low barrier to entry; however, we didn’t see it as a direct competitor, but moreso as a signal of where the category was heading than a competitor to design against.
Narrowing from hardware to software, from everyone to students.
Hardware meant slower iteration and real manufacturing cost, while software let us build and test within our sprint’s time constraints. We focused on students because it’s a life stage we’d lived ourselves, leaving us with less time to validate the problem, and more time to actually build.
Why Juno?
Starting from a concept, not a category.
Speed to Build
Software, not hardware, buildable within a 10-week sprint.
Sharper problem-fit
Validated through 50 student interviews, not assumed from market size.
A compounding moat
A knowledge graph that gets more valuable the longer it’s used, unlike a one-off tool.
Prototyping
Exploring directions for something both fun and with utility
We explored a range of concepts trying to answer to focal questions: how do you make an always-updating knowledge graph feel like something a student wants to check and not just a sprawl of data and what channels it should reach students through?
Visual Explorations
[1] Visualizing the knowledge graph directly; accurate, but unreadable at a glance.
[2] Decorated home screens personalized to users; distinct, but loses connection to Juno’s main function.
Form Explorations
[1] Passive updates on the phone’s home screen which proved to just add more noise.
[2] Just a text, arriving the moment something’s relevant.
Key user insights
Raw data isn’t actionable
Seeing your own knowledge graph doesn’t tell you anything, students needed it translated into something more than just accurate, but immediately useful.
Personalization and fun can still exist in software!
A customizable interface only felt valuable when every visual element was tied to a real function, not decoration for its own sake.
Form follows function.
As obvious as it sounds, we ended up scrapping most of the widgets above. Every visual piece that stayed had to map to a real function, and anything that was just decoration got cut as noise.
Insight 1: Raw data isn’t actionable
We translate the knowledge graph into formats students already trust like calendars and dashboards instead of exposing it directly.
Insight 2: Personalization and fun can still exist in software!
We created the room, a sort-of home away from home for college students, where every piece of furniture is a visual entry point into a real, running automation.
Designing for scale
Can AI design something reusable, or only something one-off?
Navigating our time constraints, our team thought about how we could design for every unique student scenario. For example, how could a streak tracker be built for both sleep and gym habits without having to be redesigned by and each time?
Option A:
Let AI generate a version per use case for more automations, which was faster, but inconsistent and hard to maintain.
Option B:
Hand-design a smaller set of reusable components, but each one consistent and built to scale.
We chose Option B. A streak tracker, for example, could be reused across gym habits, healthy eating, and LeetCode practice with no extra design work. The tradeoff: something like a calorie tracker, which needs its own specific inputs and logic, couldn’t be repurposed the same way. So for now, Juno’s automations skew toward what’s reusable, and highly specific, detail-heavy trackers are something we haven’t built out yet.
Designing for trust
Asking for someone’s whole life means security can’t be an afterthought.
Every integration runs through Composio, so Juno only ever accesses what a student explicitly grants. Data retrieval runs through SOC 2—compliant providers, everything is end-to-end encrypted, and our own engineers never see a user’s raw data!
This wasn’t a feature we added later because we understood if Juno is going to compound context over time, trust has to compound with it!
Reflection
My takeaways
Useful doesn’t mean constant
“Proactive” tips can easily become “annoying.” An assistant that acts before you ask has to earn the right to interrupt you, not just have the ability to. Choosing the right channels were extremely important.
Designing for scale
Choosing hand-built, reusable components over AI-generated one-offs meant fewer automations at launch, not more. Scale was about building the proper foundation early on for greater expansion later.
Outcomes
What shipped, and what’s next
As of today, Juno has 600+ people on the waitlist and 50 students actively using it in beta.
Thank you to my wonderful LavaLab team—Ronnie, Johnny, and Turat—for building with me, and to my mentors for pushing me outside my comfort zone! Grateful for this lovely community of passionate builders.
