Study Hub
Interactive study platform for three university CS courses.
Study Hub is a study platform for three of my university computer-science courses: Lenguajes Formales, Estructuras de Datos y Algoritmos, and Sistemas de Gestión de Datos. Each subject has its own dashboard of practice problems with simple, formal, and full solution tabs, topic explanations, handwritten-style study notes, and exam quizzes, all pre-authored as MDX with KaTeX math. One typed subject config and one shared design system drive all three, so a subject owns a single accent and every surface resolves to it automatically.

- Role
- Product design and engineering
- Year
- 2026
- Platform
- Software project
- Status
- in progress
The problem
CS courses lean on ideas that are genuinely dynamic: a machine consuming a string, a stack rising and falling, a parser reducing a handle. They are usually taught with static blackboard snapshots and PDFs. I wanted one place that holds a whole semester of three courses and makes those moving parts actually move, with solutions I can read at three depths and content an agent can extend without touching a database.
What I built
The work includes web, viz, content.
The choices behind it.
Solo · design + engineering
Implementation choice
Automata as first-class visualizations: DFA, NFA, PDA, and Turing-machine state graphs are auto-laid-out with ELK and rendered in ReactFlow, then a shared step runner walks an input string one symbol at a time, highlighting the active transition, growing and shrinking the PDA stack, and moving the Turing tape under the head.
Implementation choice
The whole compilers unit made interactive: FIRST and FOLLOW set computation, LL(1) and SLR parse tables, animated shift-reduce traces, and CFG grammar transforms, plus newer sorting and relational-algebra visualizers, all built on one useStepAnimation primitive that renders each frame from a real, instrumented trace.
Implementation choice
Content is pre-authored MDX, not a CMS: 39 problems and 66 solutions with simple / formal / full tabs, topic explanations, Caveat-handwritten study notes, and exam quizzes, every one with KaTeX math, where a single solved boolean in problem frontmatter is the source of truth for dashboard status.
Implementation choice
One design system across three subjects: a refined editorial-dark theme (Instrument Serif display, Instrument Sans body, Caveat notes) with a per-subject accent bridge set via [data-subject], sapphire for Lenguajes Formales, jade for EDA, amethyst for SGD, resolved once so no surface re-declares a color map.
Implementation choice
A key-free, offline MCP server: 16 typed tools (10 read, 6 write) over stdio, pure file I/O, that let Claude list and read problems, solutions, and visualizations and author new ones, with every write validated by zod schemas that mirror content.config.ts, so the server rejects exactly what an Astro build would.
Implementation choice
Genuinely runnable with zero secrets: seeded content, four self-hosted woff2 fonts, and a mocked upload endpoint mean both the live demo and a fresh clone boot with no API keys and no network fonts.
- typed MCP tools, 10 read and 6 write, key-free over stdio
- 16
- problems and 66 tabbed solutions, all pre-authored MDX
- 39
- React visualizers, from DFA graphs to database schemas
- 16
- subjects, 19 units, 0 API keys to run
- 3
Under the surface.
web
appThe Astro 6 dashboard: per-subject problem lists, simple/formal/full solution tabs, explanations, notes, and exam pages, with a React island mounted for every interactive visualization.
viz
libraryThe React visualization library: ReactFlow + ELK automata, PDA, and Turing-machine graphs, LL(1)/SLR parse tables, shift-reduce and relational-algebra traces, sorting steppers, and one shared step-animation runner.
content
contentThe MDX content store: problems, three-depth solutions, explanations, handwritten notes, and tests per subject, plus the JSON datasets that feed each visualization.
mcp
serverThe key-free stdio MCP server: 16 tools (10 read, 6 write) over pure file I/O that read and author the content store, with zod validation mirroring the Astro content config.
config
configThe subject registry: one typed SubjectConfig per course (units, topics, exams, viz enums) that drives routing, dashboards, and the per-subject accent triad.
Build details Stack, scope, and implementation
Built with
- TypeScript
- Astro 6
- React 19
- Tailwind CSS v4
- MDX
- KaTeX
- ReactFlow
- ELK.js
- Framer Motion
- Zustand
- Model Context Protocol
- Zod
- Node 22
- pnpm
- Vercel
Areas
- study platform
- automata theory
- interactive viz
- MCP
- MDX content
- offline-first
More about the project
The part that matters is that the hard theory is interactive, not static. Automata render as live state graphs: DFA, NFA, PDA, and Turing machines laid out by ELK and drawn with ReactFlow, that you step through one symbol at a time while the active transition lights up, the stack grows, and the tape moves. The parsing unit adds FIRST and FOLLOW sets, LL(1) and SLR tables, and shift-reduce traces, alongside grammar transforms and newer sorting and database-schema visualizers. A key-free MCP server exposes the whole content store to Claude over stdio as typed read and authoring tools, and the app runs fully offline with seeded content, self-hosted fonts, and no API keys.
Try the product.
Interactive demo · Open full screen