// system architecture
There is no course.
There is no content library.
Every session is made for you,
right now.
→ generated live · personalized to your role · informed by your history · never repeated
Day01.AI doesn't serve you content. It assembles it — from scratch — every time you start a session. Your role, your phase, your quiz history, your skill file. Then it picks the right AI and builds something that has never existed before.
// under the hood
Five ideas that make this
a different category of thing.
Not a feature list. The actual architecture — why it works, why it compounds, and why your data never leaves your control.
01 / THE PROBLEM
01 / 05
Every tool ships the same lesson to everyone.
The content was written before you existed. Your role, your stack, your pace, your gaps — none of it was in the room when it was authored.
It's not bad content. It's just not for you.
02 / LEARNING DNA
02 / 05
We build a model of you. It lives in a file.
learning-dna.md — a structured snapshot of how you think, what you know, and where you stall. Versioned. Yours. Not a profile buried in an analytics warehouse.
v1 after onboarding. A new version after every session.
03 / RUNTIME ASSEMBLY
03 / 05
Nothing is pre-made. Every lesson is compiled.
When you press Start, we read your DNA, pick the next concept, and assemble a lesson that has never existed before — tuned to your role, your goal, and your last mistake.
8 inputs · under 10 seconds · one lesson written for one person.
04 / THE LOOP
04 / 05
Every session teaches the system more about you.
Your answers, your hesitations, the exercises you skipped — all of it rewrites the DNA. The next session starts smarter than the last. Then the one after that.
No two users get the same session. No user gets the same session twice.
05 / ZERO KNOWLEDGE
05 / 05
Your API key never touches our infrastructure.
You bring your own Anthropic key. It lives encrypted in your browser. Every generation call goes direct from your browser to Anthropic — we are not in the middle, not in a log, not in a database.
This isn't a privacy policy. It's an architectural guarantee.
// why this doesn't exist anywhere else
Personalized, real-time generation at this depth
is an unsolved problem in learning.
We solved it.
Most "personalized" learning platforms assign you to a track. They pick a pre-made course from a library and call it your path. The content was written once, for everyone. It doesn't know you. It knows your category.
Day01.AI generates your lesson the moment you ask for it, using a context window assembled from: your exact role and specialization, your current phase in a 25–30 topic curriculum built specifically for you, your quiz performance across every past session, your exercise responses, your streak, and a living document that updates itself to track your entire AI journey. No two users get the same lesson. No user gets the same lesson twice.
pre-made lessons
everything is generated live
inputs assembled
before one word is written
learning phases
gated by real performance
seconds to generate
your personalized session
// the context assembly
Before a single word is generated,
the system reads eight things about you.
This isn't prompt-stuffing. Each input shapes a different part of the output. Your role determines the angle. Your phase determines the depth. Your quiz history tells the system where you struggled. Your skill file tells it what you already know.
Your Role + Specialization
Senior PM · Fintech · B2B SaaS
Experience Level
7 years · mid → senior transition
Current Phase
Phase 2 — Role-Specific (Day 12)
Today's Curriculum Topic
AI tools for product managers
Last 3 Session Summaries
Quiz: 80%, 60%, 100% · all completed
Quiz Performance History
Avg 78% · weak: prompt engineering
Your Living Skill File
v13 · 4,200 chars · 12 topics learned
Knowledge Base Context
Latest AI tools, model facts, updates
AI Generation
Engine
Output — generated once, for you
→ personalized · never repeated
// the memory layer
Most apps have no memory.
Yours does. And it updates itself.
When you complete onboarding, the system generates a 25–30 topic curriculum personalized to your role, industry, and experience level. That curriculum lives in a markdown document called your Skill File. After every session, the Skill File is automatically updated — the topic is marked learned, your quiz performance is recorded, your patterns are noted. The next session reads the updated file before generating anything.
This is not a progress tracker. It's a knowledge graph of you, written in plain text, versioned, and growing.
The loop that makes the system smarter with every session
// system architecture
Six systems. All connected.
All working before you see a word.
Nothing is standalone. Your Learning DNA shapes your curriculum. Your curriculum shapes your sessions. Your sessions update your curriculum. Your mentor knows all of it. The score determines what comes next.
hover any node to see its role
// the ai routing layer
Not a ChatGPT wrapper.
A task-aware, multi-provider,
self-healing AI layer.
Day01.AI works with whatever API key you bring — Anthropic Claude, OpenAI GPT, Google Gemini, or any combination. It doesn't treat them as interchangeable. Each task is routed to the right tier. If a model fails — rate limit, timeout, server error — the system automatically tries the next model in the chain. You never see the failure.
// automatic fallback — live demo
Click a model to simulate a failure.
// security
Your API key's complete journey:
Your browser → AES-256 → localStorage → Provider.
Day01.AI is not in that chain.
We don't proxy your API calls. We don't store your keys. We don't see them. When you add a key, it's encrypted in your browser using SubtleCrypto — a W3C standard, browser-native cryptography API — derived from your user ID and a per-user random salt. When you start a session, it's decrypted in memory only long enough to make the call, which goes directly from your browser to the provider. Our servers are never involved.
You type key
Plain text, in memory only
AES-256 encrypt
SubtleCrypto (W3C standard)
localStorage
Per-user salt — only you can decrypt
Decrypt in-memory
Only when making a call
Provider API
Anthropic / OpenAI / Gemini
Day01.AI Server
not in the chain · never sees your key
Zero hops through us. Your key's path ends at the provider, not us.
// readiness scoring
A score that reflects real learning,
not just completion.
AI Readiness Score isn't a progress bar. It factors in quiz accuracy, exercise depth, streak consistency, and the difficulty of your current phase. It decays if you go inactive — because readiness without practice isn't readiness. It gates phase progression: you can't advance to Role-Specific content until your quiz accuracy clears a threshold. The system respects the learning process.
Perfect quiz
+3
+15 XP
Partial quiz (≥50%)
+2
+10 XP
Exercise (>150 chars)
+3
+15 XP
7-day streak
+0.5
+5 XP
3+ days inactive
−1/day
max −10
Phase multipliers — XP scales as you advance
Phase gate: requires ≥40% quiz avg across last 5 sessions to advance. The system holds you until you're ready.
You see a 10-minute lesson.
This is what we see.
Thirty-seven decisions. Eight inputs assembled. A curriculum updated. A score recalculated. A mentor briefed on everything that happened today. And tomorrow, when you come back, the system will have been thinking about what comes next.
There's nothing else like it. That's not a claim. That's a consequence of how it's built.