Case · estud.ia · 3-minute read
estud.ia is an education platform where AI helps each student without replacing the teacher. I designed and built it end to end, and every decision below is one I made on purpose.
Website: estudia.solutions · The app: app.estudia.solutions
Each role sees its own view of the same school. Teacher, student, parent and principal are described through the personas used to explain the product.
For students, the AI changes how the teacher’s help is presented, not what it says.
The teacher writes each level. Using help costs 5, 10 or 15 points, so students try first.
Adoption is progressive, from setting up the school to using it in class. Leadership gets an institutional view without turning the system into a ranking.
They run against PostgreSQL, not SQLite, so schema and migration problems show up early.
Each gate has to pass before a change moves on. If one fails, the change stops there and gets fixed.
Why it exists
Learning is in a quiet crisis: technology moves faster than pedagogy, and AI, meant to extend human thinking, is often used as a cognitive shortcut.
4 in 5
sixth graders in Latin America and the Caribbean would not reach the minimum level of reading comprehension.
UNICEF
78%
of class time is actual teaching; the rest goes to order and administrative tasks.
OECD, TALIS 2018
11%
of student submissions show AI-generated text.
Turnitin (2024), via Education Week
Make personalization possible at scale without replacing the teacher. The teacher sets the pedagogical frame; the AI adapts how each student takes in the content and never hands over the answer. Two users drive the design: teachers, who need to personalize without multiplying their workload, and students, who need consistent support in the way they learn best.
What it shows Problem framing: starting from evidence and one hard constraint (assist, never replace) before choosing the technology.
Principle zero
In the process
The security review is step one of every commit, before the tests run.
In the product
Consent comes before access. Personal data is encrypted from the first write.
In the AI
Safety rules for minors lead every student- and parent-facing AI call, ahead of the teacher’s content.
How it was designed
The design phase ran with teachers and school leaders from different schools. They have different needs, and the work was to turn that input into one coherent product.
What it shows Stakeholder management: gathering requirements across institutions and aligning them into one product.
Decisions that shaped it
The teacher writes the help in three levels. For students, the AI only reshapes it for how they learn: visual, analytical, audio or spatial.
What it shows Product judgment: I scoped the AI to what earns trust, not to what looks flashy.
A student cannot enter without valid parental consent, and a school cannot go live until its parents have signed. Personal data is stored encrypted.
What it shows Security and compliance thinking built into the flow from day one, not bolted on.
Chat and help calls pass through a single client: safety rules in, moderation and audit out. The provider (Anthropic, OpenAI or an offline fallback) can be swapped per school.
What it shows Architecture: one choke point makes the provider swappable and every call governable.
Quality gates run before every commit: an import check, schema and service integrity, a security review and the full test suite. Continuous integration then reruns the tests on PostgreSQL.
What it shows Delivery discipline: automated gates make quality the default, not an afterthought.
Safeguards