convez.ai
PluginAutomaçãoAvançado
jimyeong

Lidless Oracle

jimyeong/lidless-oracle

Chatbot de risco de mofo com sensores IoT e Claude multilíngue

Ver no GitHub
InicialPluginScore · de 0 a 100

Chatbot multilíngue integrado com Claude e dados reais de sensores IoT para avaliação de risco de mofo em residências do Reino Unido. Fornece aconselhamento conversacional e personalizado sobre prevenção e mitigação de mofo, com suporte a múltiplos idiomas e cálculos baseados na fórmula de Magnus.

Instalação

$npx convez-ai jimyeong/lidless-oracle

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 6 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Avaliação de risco de mofo em habitações usando dados de temperatura e umidade em tempo real
  • Suporte multilíngue para comunidades diversas em habitação social
  • Integração de dados de sensores Zigbee com raciocínio de IA para aconselhamento contextualizado

Veredito

Projeto robusto que exemplifica bem a orquestração de Claude com dados IoT reais e múltiplas linguagens. A arquitetura multi-serviço (Node + Python) é sólida, embora ainda em desenvolvimento ativo com funcionalidades como persistência de conversas e alertas em tempo real por vir.

README

A tradução para o português deste README ainda está na fila.

Lidless Oracle 🔮

A multilingual LLM chatbot for mould risk management in UK homes.

Demo

▶ Watch the demo (1 min)

What this project is

Mould affects ~4% of UK homes (English Housing Survey) and disproportionately impacts social housing tenants and non-native English speakers — those who most need accessible information often face the steepest barriers to it.

Lidless Oracle is a chatbot designed to bridge that gap. Users converse in their own language about mould — its causes, prevention, and the present risk in their home — while the system fetches real environmental data from IoT sensors and reasons over it through an LLM (Claude).

Why a chatbot, why multilingual

  • Conversational over forms: most mould advice today comes as static guides or symptom-checking forms. People don't speak in form fields.
  • Multilingual by design: UK housing reaches diverse communities. An advisor that answers in Korean, Polish, Arabic, or Urdu meets users where they are. The LLM handles this natively, without a translation layer.
  • Grounded in real data: rather than generic advice, the system queries actual sensor readings (humidity, temperature) and assesses risk via the Magnus formula in a separate Python service.

Demo (May 2026)

The current build demonstrates the core conversation loop:

  • User asks in natural language: "What was last week's mould risk?" / "이번 주 곰팡이 위험은?"
  • Claude interprets the time range, calls a backend tool that fetches real IoT data, and replies conversationally with risk score, sample size, and actionable advice.

Architecture

ComponentTechRole
Mobile clientReact Native, TypeScript, Apollo ClientChat UI, auth, GraphQL
API gatewayNode.js, Apollo Server, PrismaLLM orchestration, tool routing
LLMAnthropic Claude (tool_use)Natural language understanding, multilingual response
Mould risk enginePythonMagnus formula on Zigbee sensor data
AuthFirebase Authentication + Admin SDKGoogle Sign-In, ID token verification
DatabasePostgreSQLUsers, conversations, messages

Status & roadmap

🚧 Active personal development.

Done

  • ✅ Google authentication via Firebase
  • ✅ Conversational chat UI
  • ✅ LLM tool use loop (Claude ↔ mould risk service)
  • ✅ Real IoT data integration (4,000+ sensor readings)

Next

  • ⏳ Conversation persistence per user (separate sessions)
  • ⏳ RAG layer with UK housing health guidance (NHS, HHSRS Cat 1 hazards)
  • ⏳ Real-time alerts when risk crosses thresholds
  • ⏳ iOS support

Notable design decisions

  • Multilingual via LLM, not translation layer: Claude responds in the user's language without an explicit i18n system. The system prompt sets language-matching behaviour; output quality is the LLM's responsibility.
  • Firebase Auth + own DB: Firebase handles OAuth and token lifecycle; user-domain data lives in Postgres via firebase_uid foreign reference. Backend verifies tokens with Firebase Admin SDK on every request.
  • Tool use loop: The chat resolver maintains a loop that handles arbitrary numbers of tool calls per turn, with an iteration limit for safety.
  • Multi-service backend: LLM orchestration (Node) is separate from sensor data processing (Python) — each in the language best suited to its job.

Repo scope

This repository contains the React Native client. The backend (lidless-hermes) and IoT subsystem (lidless-controller) are private repositories.


Built solo, May 2026.

Ver o original no GitHub

Bom para

Relacionados