AWS Builder Center

StudyOS

It is a personalized learning OS that eliminates context-switching for engineering students by letting them draw a rectangle around any confusing code or text on any webpage to instantly receive a grounded AI explanation, then saves the Q&A into a spaced-repetition review queue that turns momentary confusion into long-term retention.

StudyOS: The Learning OS That Finally Lets Students Ask Questions Without Breaking Flow




The Problem Every Engineering Student Knows

You're deep in a LeetCode problem. The binary search condition reads  while (low <= high)  and something feels off. Is it  <=  or  < ? Why does it matter?
What happens next is familiar to every engineering student: you copy the snippet, open a new tab, navigate to ChatGPT or Stack Overflow, paste the code, type "why  <=  instead of  <  here?", wait for the answer, copy the explanation, switch back to LeetCode, try to re-orient yourself in the problem—and somewhere in that 3-minute context switch, you've lost the thread. The question vanishes. The insight evaporates. Tomorrow you'll have the same confusion again.
This is the hidden tax of modern learning: the context-switching tax. Students don't lack resources—they drown in them. YouTube, LeetCode, GitHub, GeeksforGeeks, documentation, Discord, Stack Overflow, ChatGPT, Claude. The information is everywhere. What's missing is execution: a system that meets you exactly where you're stuck, answers exactly what you're confused about, and ensures the insight sticks.



Why Now: The Cost of Fragmentation Is Rising

We're not in an information scarcity era anymore. We're in an attention fragmentation crisis.
The half-life of technical knowledge is shrinking. Frameworks obsolete in months. New frameworks emerge weekly. The "best path" for a student's goal shifts quarterly. In this environment, the cost of learning the wrong thing—or learning the right thing in the wrong order—compounds exponentially. A student who spends two weeks learning a deprecated pattern doesn't just lose two weeks; they lose the compounding progress they'd have made on the right path.
Meanwhile, AI assistants have made getting answers trivial—but they've made grounded, contextual, verifiable answers harder to distinguish from hallucination. Students don't need more answers. They need the right answer for their exact context, right now, without derailing their flow.
StudyOS exists because the gap between having resources and executing learning has never been wider—and the cost of that gap is measured in wasted hours, missed internships, and careers that plateau before they start.



What StudyOS Is: A Learning Operating System

StudyOS isn't another flashcard app or video platform. It's a learning operating system that connects the full loop:
Today → Topic → Point & Ask → Review → Evidence → Next Action

The Complete Loop

1. Today Screen — Opens to one clear next action: "Binary Search, 60 min. You've completed Arrays and Hashing. This is your next prerequisite-ready topic." No decision paralysis. No "what should I study today?"
2. Curated Resources — Each of the 14 DSA topics comes with curated resources: a video, an article, a practice set. No more YouTube rabbit holes.
3. Point & Ask (Text) — Select any text on any page → right-click → "Ask StudyOS" → get a grounded explanation from AWS Bedrock that cites your selection.
4. Spatial Point & Ask (The Breakthrough) — Press  Alt+Shift+A , draw a rectangle around anything on any webpage (LeetCode, GitHub, PDF, documentation). StudyOS deterministically resolves exactly which DOM object you circled—code block, paragraph, table cell—using pure geometry. No LLM guessing. Then Bedrock explains only that object with citations to your selection.
5. Feedback → Save to Review — Every answer gets a Helpful/Not Helpful vote. One click → "Save to Review" → enters a spaced repetition queue (Again tomorrow / Good in 3 days).
6. Review Queue — Tomorrow, the question returns. You answer again. The interval grows. The memory consolidates. The confusion becomes permanent knowledge.
6. Evidence Accumulates — Each topic tracks: resources completed, questions asked, helpful answers, reviews completed. The topic state evolves: No Evidence → Started → Developing → Reviewed. No fake mastery percentages. Real evidence.
7. Operator Dashboard — Real-time metrics: students, activation, asks, success rate, helpful %, review saves, latency, top domains/topics. Visible proof of learning at scale.



Why It's Revolutionary

1. The Resolver-First Architecture

Most AI tools shove the entire page into an LLM and hope for the best. StudyOS does the opposite: deterministic geometry first, LLM second. The spatial resolver scores every DOM element under your rectangle using pure geometry—overlap, containment, center distance, oversized-container penalty. The top candidate wins. Only then does the LLM see the resolved object. The model never guesses what you meant. It explains what you actually selected.

2. Zero Context Switch

Alt+Shift+A → draw rectangle → ask → answer → save → keep working. The side panel opens over the page. You never leave LeetCode. You never lose your place. The question never disappears.

3. Questions Become Assets, Not Waste

Every question asked becomes a review item. Every review strengthens the memory. The student builds a personal knowledge base of their actual confusions, not generic flashcards. The operator dashboard shows: "This student asked 47 questions, saved 31 to review, completed 12 reviews, helpful rate 89%." That's not engagement theater—that's learning evidence.

4. Privacy by Default, Not Afterthought

 anchors_only  privacy tier by default: only the resolved DOM object + nearby text leave the browser. No full-page screenshots. No full-page text. No URLs with query parameters. The student sees exactly what's sent:  ✓ Resolved DOM object · ✓ Nearby text · ✕ Full page · ✕ Full screenshot . Anonymous device IDs let students try before signup; history adopts on login.

5. Coverage Lite: The Second AI Feature

Paste any resource (article, video transcript, PDF text). Bedrock evaluates it against the topic's objectives: Strong / Moderate / Weak / Missing. No embeddings. No vector DB. No OpenSearch. One prompt, structured JSON, instant verdict. "This article covers lower bound well but misses upper bound entirely."



Why It Matters Now

The world doesn't need another flashcard app. It needs a system that respects the student's attention as the scarcest resource in modern learning.
Every design decision in StudyOS serves one principle: the student's flow is sacred. No modals that steal focus. No "upgrade to pro" modals. No "watch this ad" interstitials. No "sign up to continue" walls. The extension loads in 200ms. The answer streams in 800ms. The review queue opens in 120ms. The entire system is engineered to be faster than the thought "I should Google this."



What We Built (The Stack)

Layer
Technology
Why
Frontend
Next.js 16 + React 19 + Tailwind + shadcn/ui
Fast, accessible, maintainable
Auth
NextAuth v5 + Cognito + custom extension tokens
Zero custom auth code
Database
DynamoDB (7 tables, on-demand)
Zero ops, single-digit ms latency
AI
AWS Bedrock (Claude 3 Haiku/Sonnet)
Pay-per-request, no GPU management
Storage
S3 (crops, 7-day lifecycle)
Private, ephemeral, secure
Auth
Cognito + custom extension tokens
Zero custom auth code
Infra
SAM (Cognito, 7 DynamoDB tables, API Gateway, 7 Lambdas, Bedrock IAM, S3, CloudWatch)
 sam deploy  = production
Extension
MV3, side panel, Alt+Shift+A, rectangle/circle/pen
200ms load, 0 context switch
Testing: 38 unit tests (10 for the resolver alone).  pnpm exec tsc --noEmit  clean.  pnpm build  exit 0. E2E against local DynamoDB: 11/11 pages 200, auth gating verified, resolver logic verified against live DynamoDB.



Why This Could Change How Students Learn

We've all seen "AI tutors" that are just ChatGPT wrappers. StudyOS is different because it doesn't try to replace the teacher—it amplifies the student's own effort.
The spatial resolver doesn't guess. The LLM doesn't hallucinate (it only sees the resolved object). The review queue doesn't guess intervals—it uses proven spaced repetition. The Today screen doesn't guess what's next—it computes it from prerequisites and evidence.
Every component is deterministic where determinism matters, AI where intelligence adds value.
For the student: they stop managing their learning and start doing it. For the educator/operator: they see real evidence of learning, not vanity metrics. For the system: every interaction generates evidence that improves the next recommendation.



The Vision Beyond the Hackathon

StudyOS isn't a hackathon project that dies after submission. The architecture is designed for what comes next:
  • Custom topics → students build their own roadmaps
  • Quiz generation → LLM generates, deterministic scoring
  • Resource ingestion → GitHub/YouTube ingestion → automatic coverage mapping
  • Knowledge graph → semantic concept map → adaptive Today
  • Multi-tenant → orgs, cohorts, teachers, analytics
But the core will never change: the student's flow is sacred. Every feature must pass the "does this help them stay in flow?" test.



Final Thought

In a world where AI can generate infinite content, the scarce resource isn't information—it's attention that converts to understanding. StudyOS is the first system built from the ground up to protect that conversion.
For the student staring at  while (low <= high)  wondering why not  < —StudyOS doesn't just give the answer. It gives the answer in context, saves it for tomorrow, and makes sure they never have to ask the same question twice.
That's the revolution. Not AI that answers. AI that stays.


StudyOS — Know what to learn today. Understand what stops you. Remember what matters.
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article