RENATA

Gemini swarm saves thousands of Ecuadorian teachers 15+ hrs/week — curriculum-aligned plans, rubrics & inclusive adaptations in minutes.

Comment

Inspiration

My name is David Polo. I'm 46 years old. I sell textbooks to schools for a living, and I also drive Uber to make ends meet. I don't know how to code — not a single line.

But I know something most tech founders don't: I walk into schools every single day. I see what Ecuadorian teachers go through — buried under mountains of paperwork: lesson plans, rubrics, evaluations, curricular adaptations — spending 15+ hours every week on bureaucratic documentation instead of teaching. And here's what breaks my heart: the same content is used identically in every classroom, even though every school has its own vision and every teacher has their own unique way of teaching.

I believe in what AI can do. And I decided: if education needs to change, I'll start that change myself. That's how RENATA was born — not in a Silicon Valley lab, but from the frustration of a textbook salesman who knew there had to be a better way.

What it does

RENATA is a Sovereign SWARM — a multi-agent AI system with Google Gemini at its core — that eliminates the bureaucratic nightmare of teaching and replaces it with hyper-personalized, curriculum-aligned educational resources generated in minutes.

  • Traditional approach: the teacher builds every plan by hand, copying and formatting documents for hours.
  • RENATA's approach: the teacher sets up their profile once and RENATA generates their plans, materials and adaptations in minutes.

What RENATA generates:

  • Annual Plans (PCA) aligned with the Ministry's official curriculum instructivo
  • Micro-plans (PUD) with ERCA methodology (Anticipation, Construction, Consolidation) and activities aligned to the student's cognitive level (Bloom's taxonomy) with objectives in the infinitive-verb format required by the Ministry's instructivo
  • Complete ClassKits: story premise, worksheets, readings, activities, rubrics and teacher guide
  • Rubrics with achievement-level indicators · formal assessment instruments
  • UDL/DUA adaptations · real SEN/NEE adaptations (ADHD, dyslexia, intellectual disabilities)
  • Reading plans · academic theater scripts · DECE counseling reports
  • Complete class materials personalized to each teacher

The secret weapon: the teacher's digital clone. RENATA creates a digital clone of each teacher from their profile — their subject, their level, their institution's vision, their teaching philosophy — and continuously learns from every teacher action: from their approvals, edits and preferences. Every resource is born from that clone: it doesn't sound like a robot, it sounds like that teacher wrote it — and it knows them better every time.

How we built it

Even though I'm not a programmer, I understood the architecture RENATA needed because I understand the problem deeply. Working with Google's AI-assisted development agent (Antigravity), I built the core of the platform, and refined the final development in VS Code with an AI coding assistant.

Frontend (Next.js on Vercel): a productivity dashboard, not a chatbot. Bilingual (Spanish/English).

Backend — the Swarm (Node.js/TypeScript): an asynchronous multi-agent orchestration system that processes PCA, PUD and ClassKit through a work queue.

Agent Role
Orchestrator Routes jobs and validates security (anti-prompt-injection, PII protection)
Curricular Agent RAG with 2,881 official curriculum codes
Book Agent Maps the teacher's textbook to the curriculum: provides the concrete material for each week
Identity Agent Injects the Teacher DNA
Fiscal Agent Audits every ClassKit looking for errors
Curricular Supervisor Expert-collapse that verifies PCA and PUD
Lince Chaos Red team: attacks RENATA (prompt injection, XSS) to find vulnerabilities
Fenix Retries failed generations
Economist Controls API costs in real time

AI Engine: Gemini at the core — 3.5 Flash (generation), 3.7 Flash (curriculum extraction), 3.1 Flash Lite — with a DeepSeek cost layer for the highest-volume creative step. Structured JSON prompts: sub-cent cost per generation.

Knowledge Base (RAG): 2,881 official curriculum codes (1,807 DCDs + 375 evaluation criteria + 699 indicators), verified 100% against the Ministry's official PDFs — zero-hallucination generation.

Database: Supabase (PostgreSQL) with row-level security, multi-tenant.

Challenges we ran into

  1. I don't know how to code — my biggest challenge and my greatest asset.
  2. Curriculum complexity — thousands of interconnected codes across 10+ subjects; 2,881 verified one by one.
  3. Making AI sound like an Ecuadorian teacher — the Teacher DNA solved it.
  4. Real inclusion — actionable adaptations, not checkboxes.
  5. Agent coordination — strict orchestration protocols, jurisdiction validation and output contracts.
  6. Security — Lince Chaos attacks RENATA; every ClassKit passes an automatic quality audit before reaching the teacher.

Accomplishments that we're proud of

  • A plan that takes 3–4 hours is generated in under 5 minutes
  • 2,881 official codes verified 100% — zero hallucinations
  • Real inclusion, not generic disclaimers
  • Sovereign Gemini-first architecture — no OpenAI dependency
  • Built by a non-programmer who understands education from the ground up
  • The Teacher DNA makes every output sound like that teacher

What we learned

  • Domain knowledge beats technical knowledge.
  • Teachers don't want AI content: they want their content, faster.
  • Inclusion must be engineered, not appended.
  • Multi-agent systems are the future: a single prompt can't handle curriculum alignment + personalization + inclusion + quality verification at once.
  • Gemini Flash models are remarkably powerful for structured educational content at a fraction of the cost.

What's next for RENATA

Phase 2 — The Student Revolution: Google Classroom · WhatsApp for parents · ultra-personalization of classes for each student (content adapted to every child's level, interests and learning style) · student personality engine · per-student licenses · scale to Latin America.

The vision: millions of children with an AI that doesn't replace the teacher — it empowers them — and personalizes education for every student, not just every classroom.

Built by a textbook salesman who believed education deserves better. Powered by Google Gemini.

Built With

Share this project:

Updates