Inspiration
In July 2026, the Brazilian Federal Court declared a "state of unconstitutional things" for the Maxakali people. Four times more children die there than the national average — not because of a war, but because they don't have clean water.
Brazil has 5,570 municipalities. Most have no lab, no sensors, no data. When a river is contaminated, people find out weeks later — if ever. Meanwhile, Brazil exports raw cellulose, raw soy, raw iron. We export what we have. We don't transform it.
Estes inverts this logic. We turn local waste into local solutions.
What it does
Estes is an open-source platform that helps anyone investigate what science can do with the materials they already have.
You bring two things: a local material (eucalyptus, coffee husk, rice straw, açaí) and an environmental problem (contaminated water, degraded soil, fire). The platform:
- Searches 250 million scientific papers (Semantic Scholar, Crossref, OpenAlex, Tavily)
- Extracts evidence with DOI, level of evidence, and limitations
- Generates a reproducible protocol in 3 difficulty tiers
- Lets you test, measure, and document
- Generates a legal complaint if the case is serious
It's not an app about water. It's a search engine for solutions.
How we built it
Backend: FastAPI + PostgreSQL + pgvector + Redis. AI cascade (Groq → Gemini → Ollama) with circuit breaker. Scientific engine with 4 APIs. NASA FIRMS for live fire data. Open-Meteo for climate.
Frontend: Next.js 16 + TypeScript + Tailwind + GSAP ScrollTrigger. Story-scroll narrative. Interactive 3D-style canvas particles.
Hardware (Case #01): Eucalyptus leaves → carbon quantum dots (CQDs) → fluorescent sensor strip + CNC/biochar filter. Total cost: R$138.
AI: Three models in cascade — never fails, always responds. Vision via Gemini for photo analysis. Scientific evidence extraction with anti-hallucination rules.
Challenges
- Time: 11 days from idea to submission.
- Rate limits: Built a circuit breaker to survive Groq/Gemini 429s.
- AI hallucination: Every claim must have a literal excerpt from the paper, verified against the source. If we can't verify, we say so.
- Language: Global platform, but content mostly in Portuguese. We added i18n (PT/EN) and AI responds in user's language.
- Solo team: One person + AI. No hardware lab. No funding. Just curiosity and urgency.
What we learned
- Science is accessible, but it's not translated. The biggest barrier isn't knowledge — it's language.
- People don't need a lab. They need a method.
- Open source is a business model.
- Communities don't need us to save them. They need tools to defend themselves.
What's next
- Scientific preprint (October 2026)
- Pilot in 3 Brazilian communities
- Atlas expansion to 5 Latin American countries
- Native mobile app
- Indigenous language support (Yanomami, Tikmũ'ũn, Guarani)
Hackathon disclosure
What was done BEFORE NextStep Hacks 2026
- Ideation and scope definition (September 9, 2026)
- Initial Atlas with the eucalyptus case (September 10, 2026)
- Patent search and initial technical document (September 10, 2026)
- Concept of the platform + Brazilian Atlas (September 11, 2026)
What was done DURING NextStep Hacks 2026
- Complete backend: FastAPI + PostgreSQL + pgvector + Redis
- AI cascade: Groq → Gemini → Ollama with circuit breaker
- Scientific engine: Semantic Scholar + Crossref + OpenAlex + Tavily
- 19 cases in the Atlas (9 categories, 5 Brazilian regions)
- Frontend: Next.js 16 + Tailwind + Leaflet + Framer Motion + GSAP
- Real-time data: Open-Meteo (climate) + NASA FIRMS (fires)
- Contribution module: post, vote, comment, curate
- Accessibility: TTS + high contrast
- Cultural narratives: Maxakali, Waimiri Atroari, defenders
- Story-scroll narrative with 6 scenes
- Hardware prototype: CQD sensor
Estes — Solutions that grow from the root.
Built With
- crossref
- docker
- fastapi
- framer-motion
- google-gemini
- groq
- gsap
- langchain
- leaflet.js
- nasa-firms
- nextjs
- ollama
- open-meteo
- opencv
- pgvector
- postgresql
- python
- react
- redis
- semantic-scholar
- sentence-transformers
- tailwindcss
- tavily
- typescript
Log in or sign up for Devpost to join the conversation.