convez.ai
AgenteDados & AnalyticsIntermediário
barkhajava

Manufacturing QA Assistant

barkhajava/manufacturing-qa-assistant

Agent que consulta dados de qualidade manufatureira em linguagem natural

Ver no GitHub
InicialPluginScore · de 0 a 100

Agent LangChain que responde perguntas em linguagem natural sobre dados de qualidade em manufatura. Integra Claude com SQLite para análise de linhas de produção, taxas de defeitos e tendências — sem serviços externos, tudo rodando localmente.

Instalação

$npx convez-ai barkhajava/manufacturing-qa-assistant

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

Casos de uso

  • Análise rápida de taxas de defeito por linha de produção
  • Comparação de performance entre linhas em períodos específicos
  • Investigação de tendências de qualidade ao longo do tempo

Veredito

Implementação sólida de agent LangChain para contexto manufatureiro real. Demonstra padrões production-ready (tool-calling, FastAPI, SQLite) mas ainda com dados mock — útil como template para empresas que precisem de similar em seu próprio stack.

README

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

Manufacturing QA Assistant

A LangChain agent that answers plain-English questions about manufacturing quality data. Ask things like "Which line had the highest defect rate last week?" and get a direct, data-grounded answer.

Screenshots

HomeThinkingResponse
Home screen with suggested questionsAgent thinking stateData-grounded answer

Why I built this

I've spent 5+ years building production ML systems — but agent frameworks like LangChain were missing from my toolkit. This project is my hands-on dive into agentic LLM patterns, tool-calling, and the MLOps realities of shipping LLM apps.

Stack

LayerTech
BackendPython 3.11, FastAPI
AgentLangChain + Anthropic Claude (claude-sonnet-4-5)
DatabaseSQLite (stdlib sqlite3)
FrontendSingle HTML file, vanilla JS
Configpython-dotenv

Project layout

app/
  main.py        – FastAPI app, /chat endpoint, serves UI
  agent.py       – LangChain agent setup + run logic
  tools.py       – 3 @tool-decorated functions
  db.py          – SQLite helpers
  static/
    index.html   – Chat UI
scripts/
  seed_data.py   – Generates mock data into SQLite
data/            – SQLite file lives here (gitignored)

Quick start

# 1. Add your Anthropic API key
cp .env.example .env
# open .env and set ANTHROPIC_API_KEY

# 2. Run everything in one command
./start.sh

start.sh creates the virtualenv, installs dependencies, seeds the database (90 days × 4 lines of mock data — only on first run), and starts the server.

Open http://localhost:8000 in your browser.

How it works

┌──────────────┐
│   Browser    │  HTML + vanilla JS
└──────┬───────┘
       │ POST /chat
       ↓
┌──────────────┐
│   FastAPI    │
└──────┬───────┘
       │
       ↓
┌──────────────────────────────────────┐
│      LangChain AgentExecutor         │
│  (Claude decides which tool to call) │
└──┬─────────────┬───────────────┬─────┘
   ↓             ↓               ↓
 list_lines  query_metrics  get_defect_trends
   └─────────────┴───────────────┘
                 ↓
            ┌─────────┐
            │ SQLite  │
            └─────────┘

The three tools are plain Python functions registered with @tool. LangChain reads their docstrings to decide which one(s) to call for a given question. No external services — everything runs locally.

Example questions to try

  • "List all production lines"
  • "How is Line B trending over the last 4 weeks?"
  • "What was the defect rate on Line C six weeks ago?"
  • "Which line had the most defects this week?"
  • "Compare Line A and Line D for the past month"

Limitations

  • Mock data only — real integration with production systems would require a proper data pipeline and auth layer
  • No eval harness yet (planned) — agent responses are verified manually
  • Single-user, stateless between page refreshes

Get your Anthropic API key

https://console.anthropic.com/

Ver o original no GitHub

Bom para

Relacionados