Sestara

Access to private tutoring predicts exam outcomes more than ability. That's an infrastructure failure. Sestara fixes it - AI curricula, mastery tracking, 20+ languages, free

Comment

Sestara

Inspiration

Chiara studied 9 hours a day for six months.

Colour-coded binders. Sticky notes on every wall. Three different prep books. Zero social life. She was preparing for the test di ammissione a medicina — Italy's single-shot university medicine entrance exam. One attempt. One day. Six months of your life on the line.

She failed by 4 points.

Not because she wasn't smart. Not because she didn't work hard enough. Because she spent weeks on topics that were barely tested and skipped entire chapters she assumed she knew. She had no feedback loop. No way to distinguish real mastery from false familiarity. No system that could tell her you're studying the wrong things.

She studied everything. When she needed to study the right things.

And here is what makes this more than one student's story: the ones who passed weren't smarter than Chiara. Many of them had private tutors — professionals who knew the exact weighting of every topic, who caught every blind spot before it became a 4-point gap. That knowledge existed. It just wasn't available to her.

That's an institutional failure. SDG 16 calls for strong institutions built on equity and access — systems that don't quietly filter outcomes by the size of your family's tutoring budget. A university entrance system where your result depends on whether you can afford a €50/hour coach isn't strong. It's broken. And education is the infrastructure that determines everything else — which careers are reachable, which cities you can live in, which futures are yours to take. When that infrastructure fails, SDG 11's promise of inclusive, opportunity-rich communities fails with it.

We couldn't fix what happened to Chiara. But we could make sure the next student — in Italy, in India, in Brazil, in any language, facing any high-stakes exam — never walks into that room underprepared again.

That's Sestara. Named after sestante — the Italian word for sextant. The instrument sailors used to navigate open ocean with nothing but stars and math. We think learning should feel like that: oriented, purposeful, and never lost.

What it does

Sestara is an AI-powered learning platform that turns any goal into a structured, trackable, and emotionally supported study journey — with the precision of a private tutor and the accessibility of a free tool.

It is also, practically speaking, educational infrastructure for students who don't have access to institutional support. Every feature exists to close the gap between a student who can afford expert guidance and one who cannot.

🗺️ Exam Curriculum Engine

This is the core of Sestara and nothing else in edtech does it quite like this.

Type your exam. Sestara generates the complete, specific curriculum for it — every argument you need to know, broken into sections, each with estimated study hours, dedicated quizzes, and a live completion percentage. Not a generic study plan. Not a list of topics pulled from the internet. The exact knowledge map for your exact exam, structured the way the exam actually tests it.

You open Sestara knowing you need to pass the test di ammissione a medicina in 6 months. Sestara tells you: here are your macro-topics, here are the sections inside each one, here is how many hours each realistically takes, here is where you stand right now. The kind of plan a private tutor with ten years of exam experience would hand you — generated in seconds, updated in real time as you progress.

The 6-month Italian medicine semester structure is built in. So are exams across multiple countries and languages, because Chiara is Italian but the problem is universal.

📊 Topic-Level Progress Tracking

Your dashboard is your command center. Active roadmaps, topics completed vs. remaining, overall progress percentage, weekly study time, streaks. You always know exactly what you've mastered and what still needs work. The gap between feeling prepared and being prepared becomes visible, measurable, and closeable.

✅ Built-in Quiz Engine

Generate tests instantly, directly on Sestara — no external tools, no copy-pasting, no setup. Multiple choice questions come with explanations for every answer. Not just correct or wrong, but why. Short answer mode forces active recall that goes beyond pattern-matching. Every quiz feeds back into your completion percentage, closing the loop between studying and actually knowing.

🧪 Custom Quiz Studio

For users who want full control, Sestara generates a ready-to-use prompt with a structured JSON schema. Paste it into ChatGPT, Gemini, Claude, Copilot, Llama, Perplexity — whichever you use. Paste the JSON back. Practice instantly. Your library grows with every session. AI-agnostic by design, so no student is locked out by a paywall or a provider.

Two quiz modes. One for speed, one for power. Both feeding the same progress system.

📹 Quiz from YouTube

Paste any lecture, tutorial, or explainer video and Sestara generates a quiz from it. Passive watching becomes active recall in seconds. Every video is now a study session.

🃏 Automated Flashcards

Every section of your curriculum auto-generates flashcards with spaced repetition logic. Hard cards come back sooner. Easy cards fade back. You review what you actually need to review, not what you already know by heart.

💬 StudyBuddy

An on-site AI companion — natively integrated, zero setup — that knows your full curriculum and actually cares about your progress. Ask it to explain a concept, simplify a theorem, or give you an example. But also come to it when you're stuck, frustrated, or three weeks behind schedule. StudyBuddy doesn't just answer questions. It checks in. It encourages. It recalibrates you when you're spiraling.

Under the hood, StudyBuddy is powered by Wolfram Alpha's LLM API — silently, when it matters. When a question touches computation, Wolfram's result is injected into the response as ground truth before the answer is generated. A student asking about a differential equation or a chemical equilibrium gets mathematically verified precision, not a confident approximation. When Wolfram isn't needed — or isn't available — StudyBuddy falls back to its native AI layer seamlessly. No loading states. No attribution. No difference visible to the student. Just correct answers.

The difference between a student who quits and one who pushes through is rarely intelligence. It's having someone in their corner. StudyBuddy is that someone — at 2 AM, the night before the exam, when no one else is awake.

📖 Simple Explanations

Before complexity, every topic gets a plain-language breakdown. No student should feel dumb for not knowing something yet. Simplicity is not dumbing down — it's the entry point to everything harder.

🔥 Streaks, Reminders & Important Dates

Weekly streaks, exam countdowns, and study reminders to keep momentum alive between sessions. Consistency beats intensity. Sestara is built around that truth.

How we built it

Layer Technology
UI Framework React 18 + TypeScript
Build Tool Vite
Styling Tailwind CSS + shadcn/ui (Radix UI primitives)
Animations Framer Motion
Routing React Router
Data Fetching TanStack React Query
Rich Text Tiptap
Math Rendering KaTeX
Drawing Canvas Fabric.js
Data Visualization Recharts
Schema Validation Zod
Form Management React Hook Form
SEO React Helmet Async
Backend Lovable Cloud (auth, database, edge functions, storage)
Computation Layer Wolfram Alpha LLM API (with silent Gemini fallback)
AI Layer Native integration + AI-agnostic JSON schema exchange
Compliance GDPR-aligned architecture, TLS in transit, no data resale

The architecture is modular by design. Each feature — curriculum engine, quizzes, flashcards, StudyBuddy — is a self-contained module sharing one unified backend via Lovable Cloud. This made parallel development between team members clean, fast, and conflict-free across two countries.

KaTeX ensures every mathematical formula renders with typographic precision — critical for medicine, physics, and engineering exam tracks. Fabric.js powers a drawing canvas so students can sketch diagrams and work through problems visually, not just textually. Recharts turns raw progress data into visual feedback that actually motivates. Tiptap gives StudyBuddy and Simple Explanations a rich text surface that feels like a real document, not a chat window.

Wolfram Alpha's LLM API runs as a Supabase edge function behind StudyBuddy. The AI decides when computation is needed — no hardcoded triggers, no regex rules. When Wolfram is called and succeeds, its result becomes ground truth inside the generation prompt. When it fails for any reason, the fallback is instant and invisible. Students never interact with the computation layer directly; they just get correct answers.

The dual quiz architecture was the most deliberate technical decision we made. Rather than choosing between building our own AI or depending entirely on external providers, we built both. Native integration for instant, no-setup quizzing with explanations. Custom Quiz Studio for power users who want full control with any model they already trust. Zod validates every JSON schema exchange so malformed outputs never crash a user mid-session.

Privacy and data responsibility were treated as design constraints, not afterthoughts. We are a European product serving students across 20+ languages and jurisdictions. Building with GDPR principles in mind — minimal data collection, transparent processing, no resale — shaped architectural decisions throughout development, because students sharing their learning data deserve to be treated with care. Full compliance is an active and ongoing process, not a checkbox.

Challenges we ran into

Curriculum precision at scale was the hardest problem we solved. Generating a curriculum that reflects the actual structure, actual weighting, and actual hour requirements of a specific exam — not a generic approximation — required extensive prompt engineering, validation layers, and iteration against real exam syllabi. Getting it wrong means a student studies the wrong things. We couldn't afford to get it wrong.

AI-agnostic ingestion was harder than it looked. Every AI model returns subtly different JSON structures. Building schema validation robust enough — powered by Zod — to handle malformed outputs gracefully without crashing users mid-session required significant iteration and edge case hunting.

Integrating Wolfram Alpha without breaking the experience required careful design. The temptation is to surface it — to show a badge, a loading spinner, a "verified by Wolfram" tag. We resisted all of it. The real challenge was making computation invisible: Wolfram either makes StudyBuddy more accurate or it doesn't run. The student should never know which is happening. Achieving that required a tight edge function with graceful error handling and a prompt architecture that absorbs Wolfram's output naturally rather than quoting it.

Streak logic across timezones is a deceptively nasty problem. A student studying at 11:58 PM should not lose their streak at 12:01 AM. We implemented UTC-normalized day boundaries with a grace window. Small detail, enormous difference in how the product feels.

Rendering math and diagrams correctly across exam tracks required integrating both KaTeX for formula precision and Fabric.js for visual problem-solving. Making them feel native — not bolted on — inside a fast, animated UI took careful architectural work.

Making StudyBuddy feel human without feeling fake was a genuine design challenge. Motivational AI tips quickly into cringe. We tuned extensively to find the line between warm and hollow, between encouraging and patronizing. The goal was a companion that feels real at 2 AM when you're exhausted and behind — not a chatbot with an exclamation mark problem.

Supporting 20+ languages without degrading curriculum quality was a localization challenge that went far deeper than translation. Exam structures, topic naming conventions, and pedagogical frameworks differ meaningfully across countries. A Brazilian vestibular is not a rearranged version of a British A-Level. Making each curriculum feel native to its context — not just linguistically but structurally — required building with genuine cross-cultural precision.

Accomplishments that we're proud of

  • Exam Curriculum Engine — not study plans, full exam curricula. Every argument, every section, estimated hours per topic, dedicated quizzes, live completion percentage. The most precise AI-generated study tool we have seen at this level of accessibility. A private tutor's decade of exam expertise, available to anyone with a browser.
  • Dual quiz architecture — native built-in AI for instant testing with explanations, plus AI-agnostic studio for users who want to bring their own model. Nobody is priced out. Nobody is locked in.
  • StudyBuddy with Wolfram-backed precision — not just an answer engine, a companion that keeps students in the room when everything in them wants to quit. And when the question is mathematical, the answer is verified, not guessed.
  • YouTube-to-quiz pipeline — any video becomes a test in seconds.
  • KaTeX + Fabric.js integration — students can read formulas with typographic precision and sketch diagrams to think visually. Because some problems need a pen, not a keyboard.
  • 20+ languages at launch — not a translation layer dropped on top of an English product. A core architectural decision that shaped how curricula are structured, how StudyBuddy communicates, and how every feature was designed — because the student who needs Sestara most may not speak English.
  • Privacy-conscious architecture — building with GDPR principles embedded in our data model from the design phase, not retrofitted at the end. Because students deserve to know their data isn't being traded away.
  • Built across two countries in one hackathon — Italy and India, two students, one unified product. The same collaboration Sestara is designed to enable for the students who use it.
  • A product that is genuinely usable right now — real curricula, real streaks, real progress tracking, real quizzes. Not a demo. Not a mockup. Not a pitch.

What we learned

Precision is a feature. Students don't need more content. They need to know exactly what to study, for how long, and whether it's actually working. The Exam Curriculum Engine exists because vague guidance is just noise dressed up as help. Every hour estimate and completion percentage we show is a promise we made to not waste someone's time.

Computation should be invisible. The best version of Wolfram Alpha integration is one the student never sees. They don't need to know that their answer was verified against a symbolic computation engine — they just need to trust that what Sestara tells them is correct. Building technology that disappears into the experience is harder than building technology that shows off. It's also more valuable.

Flexibility beats lock-in. Designing around both native AI and a JSON schema exchange made Sestara dramatically more resilient and democratic. A student with a free Gemini account gets the same quality experience as one with a premium subscription.

Habit design is product design. Streaks, progress rings, and weekly tracking are not features — they are the mechanism that makes users return tomorrow, and the day after, and the day before the exam.

The emotional layer is the product. Students don't quit because they run out of resources. They quit because no one tells them they're on the right track, or that they can still recover. Chiara didn't fail because she lacked information. She failed because no system ever cared about whether she was learning the right things. Building that care into StudyBuddy was the most important design decision we made.

The right tool for the right problem. KaTeX for math. Fabric.js for diagrams. Tiptap for rich explanations. Recharts for progress. Wolfram for computation. Every library in our stack exists because a student somewhere needed it — not because it was impressive to list.

Simplicity is the entry point to everything harder. Every plain-language explanation we wrote made the next difficult concept land better. We will never apologise for making things clearer.

Infrastructure doesn't have to be expensive to be real. We built Sestara with the tools available to two students across two continents. That constraint became the proof of concept: if we could build this here, students anywhere can use it. Accessible tools, when used with precision, produce institutional-grade outcomes.

Localization is not translation. Supporting 20+ languages taught us that the real work isn't converting words — it's understanding how each country structures knowledge, frames exams, and expects information to be delivered. That work made Sestara meaningfully multilingual, not just technically so.

What's next for Sestara

  • Collaborative roadmaps — study with friends, share progress, compete on leaderboards
  • Mobile app with offline flashcard sync for students who cannot always rely on connectivity — because unreliable internet access should not mean unreliable exam preparation
  • Predicted readiness score — a single number that answers "are you actually ready?" based on mastery depth, time invested, section weight, and quiz performance across the full curriculum
  • Institutional tier — for schools, coaching centers, and universities to deploy Sestara at scale with teacher dashboards and cohort analytics. This is where Sestara becomes a genuine piece of public educational infrastructure: schools in underserved communities gaining access to the same curriculum-mapping precision that elite prep programs charge thousands for
  • Expanded exam library — more countries, more languages, more high-stakes tracks built with the same depth and precision as the ones we ship today
  • Civic exam tracks — extending Sestara to citizenship tests, civil service exams, and professional licensing, directly supporting SDG 16's goal of accessible, accountable public institutions that don't exclude based on economic position
  • Full GDPR compliance roadmap — completing the compliance architecture we began designing for: Article 28-compliant DPAs, formal data subject rights flows, and audit-ready documentation. Because getting this right matters more than claiming it early.

Chiara is preparing to retake her exam.

This time, she has Sestara.

Built With

Share this project:

Updates