❯ /osis
⏺ ⠀
⠀⠀⠀⠀⠀⣠⣄⠀⠀⠀ ▲ Osis v1.11.1 · osis.dev
⠀⠀⠀⠀⣰⣿⣿⣆⠀ Build products people love, faster
⠀⠀⠀⣼⣿⣿⣿⣿⣧⠀ ⠀
⠀⢀⣼⣛⠋⠉⠁⠀⠈⢷ ⠀
⢀⣾⣿⣿⣿⣿⣿⣶⣶⣶⣷⡀ Osis Lab ·
⠈⠛⠛⠛⠛⠛⠛⠛⠛⠛⠛⠁ Saturday, April 11, 2026
⠀
⠀
Welcome to Osis 👋 Let's set up your product.
I'm going to take a quick look at what you've got here.
⏺ I'm setting up your product docs now...
⏺ Good. Now let me wire up CLAUDE.md.
⏺ ✌️ You just installed a product team directly into your codebase.
From now on, just say "osis" in any conversation and I'll think product with you.
Here's what I see:
osis-lab (product TBD)
┌──────────────────────────────────────┐
│ Presentation Layer │
│ Header · Footer · ThemeToggle │
├──────────────────────────────────────┤
│ Route Shell │
│ / (hero + features) │
│ /about (single card) │
├──────────────────────────────────────┤
│ Design Token System │
│ sea-ink · lagoon · palm · sand │
│ light/dark · glass cards · anim │
├──────────────────────────────────────┤
│ Build Pipeline │
│ Bun > Vite 8 > TanStack Start │
│ + Tailwind 4 · Vitest · ESLint │
├──────────────────────────────────────┤
│ [empty] │
│ No server functions, no data, │
│ no API, no auth, no state │
└──────────────────────────────────────┘
From there: name the product and write the manifesto, then define the v0 thesis
(what are we proving?), then ship one feature that tests that thesis.
▲ What is the one thing a user should be able to do in v0 that would tell you
whether this product is worth building?
✻ Brewed for 3m 47s
❯ this is the idea i want to build, had this while walking
You're a flash card game for learning the language of any industry you type in
what you wanna learn there's already a set that and if not agent creates one for
you and you have flashcards of the term click button for more context then can
I hold space to record your answer and then the AI will have a conversation with
you to work on your mental models press space to skip at any time turn off
⏺ Here's what I'm hearing:
A flashcard game that teaches you to speak any industry's language. You type the
domain (finance, biotech, legal, whatever), you get a deck. If one doesn't exist,
an agent builds it on the fly. The core loop is: see term → tap for context →
speak your understanding → AI coaches your mental model through conversation.
Space to skip, toggle the conversation off when you just want to drill.
A few things that make this interesting to me:
1. The voice loop is the real product. Flashcards are commodity. The moment you
hold space and explain a concept back, you're doing active recall + elaboration,
which is where actual learning happens. The AI conversation layer is what makes
this not just another Anki clone.
2. "Language of any industry" is a wedge, not a feature. You're not building a
flashcard app. You're building the fastest way to become conversational in a new
domain. That's a very different product.
3. Agent-generated decks are a bet on coverage as a moat. If anyone can type
"venture capital" or "supply chain logistics" and get a deck in seconds, you
never have a cold start problem.
Some tensions I want to surface:
- Who is this for? Someone switching careers? A founder entering a new market? A
student? The answer changes everything about depth, tone, and what "success"
looks like.
- What does "conversation" actually mean here? Is the AI correcting you ("not
quite, a cap table is actually..."), or is it Socratic ("you mentioned dilution,
how does that connect to what we covered about preferred shares?")?
- How do you know the agent-generated decks are good? A wrong definition in
finance or medicine isn't just unhelpful, it's dangerous. Is there a quality
layer?
▲ Before I write anything: who did you picture using this while you were walking?
What was the scenario in your head?
❯ yeah see the term, explain what you know about it, llm evaluate pushed back,
then there can be a go deeper button
yeah one of my best friends has been trying to transition from consulting to pe
and the advice i gave him is know every single term/lingo they use in pe, that
way when you talk to them, you can speak their language at their level. i
noticed it in my own career, when entering an industry, understanding the mental
models and being able to converse fluently at their level is a huge unlock
lets keep our example here for startups because that's what i understnad haha
im thinking first response is correcting the mental model/an evaluation, the go
deeper can be more socratic
should be voice first, but we can do offer text box as secondary
im thinking we can do a full product with auth and analytics, you create your
category/industry, track your progress
i agree with accuracy, we need a protocol/methodology with redunancy check for
this rank sources of truth etc.
yeah to answer your final question again its my good friend oliver
⏺ Good. Oliver's story is the product. Let me play it back:
Oliver is in consulting, wants to break into PE. He's smart enough to do the
work, but when he sits across from PE people, he doesn't sound like one of them.
He says "revenue" when they say "EBITDA." He says "strategy project" when they
say "deal flow." The gap isn't knowledge, it's fluency. And fluency is what gets
you taken seriously in a room.
So the product is: the fastest way to become conversationally fluent in a new
professional domain.
Not a study tool. Not a quiz app. A fluency engine.
Here's how I'd frame the core loop for Oliver:
See term: "Carried Interest"
↓
Hold space, explain what you know
↓
AI evaluates: "You've got the basics. But you said
it's a 'bonus' — in PE they'd push back on that.
Carry is a share of profits, not compensation.
The distinction matters when you're talking to LPs."
↓
[Go Deeper]
↓
Socratic: "If carry is profit-sharing, why do you think
GPs sometimes negotiate carry on a deal-by-deal basis
vs. across the whole fund? What's the incentive difference?"
That's powerful. Oliver doesn't just memorize "carried interest = 20% of
profits." He builds the mental model that lets him use it correctly in a live
conversation.
A few things I want to nail down before we start writing docs:
On the deck generation protocol. You said redundancy checks, ranked sources of
truth. I think this is critical enough that it should be a first-class system,
not an afterthought. The AI that builds decks needs to be a different agent (or
at least a different prompt) than the one that coaches. Builder cares about
accuracy. Coach cares about pedagogy. Mixing those incentives gets messy.
▲ Does that split feel right to you?
On "full product with auth and analytics." I agree, but I want to sequence it.
Oliver doesn't need a login to validate whether this thing works. He needs 20 PE
terms and the voice loop. If that conversation makes him feel sharper after 15
minutes, we have a product. Auth and progress tracking are v0 features, but
they're not the experiment. The experiment is: does the voice loop actually build
fluency?
▲ How do you want to prioritize?
On naming. "osis-lab" is the repo, but this product needs a name.
▲ Any instinct?
❯ jsut thought of a featured, we can have verified language sets by experts, can
even create a marketplace, that would be cool
yeah you nailed it! So the product is: the fastest way to become
conversationally fluent in a new
professional domain.
exactly this is all about mental models, this will really help oliver
totally agree, needs to be a first class system. this is really awesome too
because user have freewill to tell us what decks they want naturally, and we can
pick those up and get them verified. yes will be an ai workflow, ingests query
-> protocol. focus on high leverage, pareto's principle, find the 20% that gets
you 80%. users should interact with the high leverage first. so we scan for
terms, rank by leverage, create a source of truth. and feed terms as they
progress. but the goal is get user to paretos optimum asap, but with depth, its
about the mental models and speaking the language, if that makes sense
yeah we can have a trial where they get their first session free etc. basically,
how can we get them to see the value and feel improvement, will take some
tweaking to place the paywall right. but there is massive roi in this product,
if people can learn the industry and get a high paying job, the product pays for
itself many times over
⏺ The Pareto framing is the product insight that separates this from every
flashcard app. You're not giving Oliver 500 PE terms alphabetically. You're
giving him the 40 terms that let him hold his own in any PE conversation, ranked
by how often they come up and how much credibility they signal. That's a
fundamentally different experience.
And the progression model follows naturally from that: get to Pareto optimum fast
(breadth), then go deep on the mental models behind each term (depth). Oliver
feels dangerous after session one. He feels fluent after a week.
The marketplace and verified sets are a strong long-term play, but that's a v1+
vision. Parking it there.
Here's where I think we're aligned:
- Core thesis: Fluency, not memorization. Mental models, not definitions.
- Key mechanism: Pareto-ranked terms → voice-first active recall → AI evaluation
→ Socratic depth on demand.
- Deck protocol: First-class AI system. Ingests a domain query, scans for terms,
ranks by leverage, builds a source of truth. Separate from the coaching layer.
- Business model: Free first session to prove value. Paywall after they feel the
improvement. ROI framing: this product pays for itself if it helps you land the
role.
- Target user: Professionals entering a new domain (career switchers, founders,
consultants moving industries). Oliver is the archetype.
Two things before I start writing:
Name. You dodged this one. Does anything come to mind? The product is about
speaking the language, fluency, sounding like you belong.
▲ Any gut reaction?
V0 scope. I want to propose: the experiment is one domain (startups), the voice
loop, and the Pareto-ranked deck. No auth, no progress tracking, no marketplace.
Just: "type startups, get 20 terms, hold space, get coached." If Oliver feels
sharper after that session, everything else is worth building. If he doesn't, no
amount of auth or analytics saves it.
▲ Sound right?
❯ would be cool if we had a form that allowed industry experts to really quickly,
low friction evaluate a deck, that will be huge for this, bring credibility fto
the platform. this is kind of like masterclass, but useful lol
yeah we also need to dedupe decks, a cool pattern could be every user kind of
gets a unique deck based on their usage of it, but there is a core shared, the
way it get surfaced algorithmically based on usage
yeah for marketing we can have success stories of people landing high salary
jobs after using this app to prep
yes right on with target user, id add college students as well preping for
internships, basically pre professionals
yea hname, lingo,jargon idk, we want to sell the promise though which is fluency
in an industry which is career success, acutalization but appealing to
professionals
yeah totally agree with you on v0, we basically need to test, does oliver get
value from this, and iterate until the core product produces value. you are so
right
⏺ "Like Masterclass, but useful" is a great internal tagline. That expert
verification form is smart, it solves credibility and accuracy in one move. Low
friction for the expert, high value for the platform. Parking it with marketplace
as v1+ vision, but it's load-bearing for the long-term moat.
The personalized deck pattern is interesting. Shared core (the Pareto set
everyone needs), but the edges adapt based on what you struggle with, skip, or go
deep on. Every Oliver gets the same foundation, but the progression is his.
On the name: you want the promise, not the mechanism. "Lingo" and "Jargon"
describe what you're learning. You want something that describes what you become.
A few to react to:
- Fluent (direct promise, "get fluent in PE")
- Versed ("get versed in biotech")
- Parlance (the language itself, a bit cerebral)
- Conversant (literally means "knowledgeable enough to discuss intelligently,"
which is exactly the product)
No pressure to pick now. But the manifesto reads better with a name.
I think we're aligned. Here's what I want to write:
1. Manifesto: why this exists. Oliver's story. Fluency as the unlock. Mental
models over memorization.
2. V0 thesis: the hypothesis we're testing. "One domain, voice loop,
Pareto-ranked deck. Does it produce fluency?"
3. V0 product definition: the core systems (deck protocol, coaching engine, voice
loop) and how they connect.
▲ Want me to go ahead? And pick a name, even a working one. We can change it later.
❯ yeah those are great names, we can use versed for now
⏺ Good. Versed it is.
Let me write the manifesto, v0 thesis, and v0 product definition now. Back in a minute.
⏺ 3 agents finished (ctrl+o to expand)
├─ Write manifesto for Versed · 2 tool uses · 13.1k tokens
│ ⎿ Done
├─ Write v0 thesis for Versed · 2 tool uses · 13.1k tokens
│ ⎿ Done
└─ Write v0 product definition · 2 tool uses · 13.5k tokens
⎿ Done
⏺ Done. Here's what we built:
Manifesto (osis/manifesto.md): Why Versed exists. Oliver's story, the fluency
gap, five core beliefs. This is the soul of the product.
V0 Thesis (osis/v0/thesis.md): The bet we're making. One Pareto-ranked session
with voice recall + AI coaching produces real fluency. We test with startups, ~20
terms. If Oliver feels sharper after 15 minutes, everything else is worth
building.
V0 Product Definition (osis/v0/product.md): Three systems. The Deck Protocol
(accuracy-first, Pareto-ranked, separate from coaching). The Voice Loop (hold
space, explain, get evaluated, go deeper). The Presentation Layer (one term at a
time, professional, voice-first).
Twin (osis/twin.md): Where the codebase actually is today. Fresh scaffold, no
product logic yet, strong design token foundation.
The docs are ready. The next natural step is the v0 strategy (how do we sequence
the build?) and then an implementation plan for the first phase. But that's only
worth doing when you're ready to code.
▲ What's on your mind?
