convez.ai
SkillBusca & PesquisaAvançado
JuniorRaja

MeGPT — Gêmeo Digital Pessoal com RAG

JuniorRaja/MeGPT

Chatbot RAG full-stack para criar seu assistente IA pessoal com histórico

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill de chatbot RAG em produção que cria um gêmeo digital pessoal capaz de responder perguntas sobre trabalho, projetos, habilidades e interesses de uma pessoa. Integra Qdrant para busca vetorial, FastAPI no backend, Next.js no frontend e suporta múltiplos LLMs via LiteLLM com rastreamento em Langfuse.

Instalação

$npx convez-ai JuniorRaja/MeGPT

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 1 mês
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Criar portfólio IA interativo que responde sobre você automaticamente
  • Implementar chatbot corporativo baseado em knowledge base pessoal ou empresarial
  • Prototipo de busca semântica com memória de conversação em sessões

Veredito

Projeto maduro e bem documentado para quem quer montar RAG production-ready com pilha moderna (FastAPI, Next.js, Qdrant). Código limpo e configurável via env vars, mas exige familiaridade com orquestração Docker e múltiplas APIs (Groq, OpenAI, Anthropic).

README

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

MeGPT — Digital Twin of Prasanna R

A production-grade personal AI twin that answers questions about Prasanna R (PR) — his work, skills, projects, opinions, travel, and interests. Powered by RAG over a personal knowledge base with a Claude-style UI.


What is this?

MeGPT is a RAG-based chatbot that:

  • Embeds and stores PR's personal knowledge (bios, projects, writing, talks) in Qdrant
  • Retrieves relevant context for each query and feeds it to an LLM via LiteLLM
  • Maintains conversation sessions in PocketBase
  • Traces every request in Langfuse for observability
  • Serves a polished Next.js UI with 3 visual themes (Claude, GPT, Grok)

Tech Stack

LayerTechnology
FrontendNext.js 14 App Router + Tailwind
BackendFastAPI (Python 3.11)
Vector DBQdrant
LLM ProxyLiteLLM
TracingLangfuse
StoragePocketBase
Embeddingsfastembed (nomic-embed-text-v1.5)
InfraDocker Compose

Local Setup

Prerequisites

  • Docker & Docker Compose
  • Node.js 20+
  • Python 3.11+

1. Clone and configure

git clone https://github.com/JuniorRaja/MeGPT
cd MeGPT
cp .env.example .env
# Edit .env — fill in GROQ_API_KEY at minimum

2. Start infrastructure

docker compose up -d qdrant litellm pocketbase

3. Start the API

cd api
python -m venv venv
source venv/bin/activate   # Windows: venv\Scripts\activate
pip install -r requirements.txt
uvicorn main:app --reload --port 8000

4. Start the UI

cd ui
cp .env.local.example .env.local
npm install
npm run dev

Visit http://localhost:3000

5. Ingest your knowledge

curl -X POST http://localhost:8000/ingest \
  -H "Content-Type: application/json" \
  -d '{"text": "Prasanna R is a full-stack engineer and product builder...", "source": "bio", "category": "about"}'

Environment Variables

VariableDescription
GROQ_API_KEYGroq API key (primary LLM)
OPENAI_API_KEYOpenAI key (deep/gpt-4o model)
ANTHROPIC_API_KEYAnthropic key (smart/claude model)
LITELLM_MASTER_KEYLiteLLM proxy auth key
LANGFUSE_PUBLIC_KEYLangfuse Cloud public key
LANGFUSE_SECRET_KEYLangfuse Cloud secret key
LANGFUSE_HOSTLangfuse host (default: cloud.langfuse.com)
POCKETBASE_URLPocketBase server URL
POCKETBASE_ADMIN_EMAILPocketBase admin email
POCKETBASE_ADMIN_PASSWORDPocketBase admin password
QDRANT_URLQdrant server URL
EMBED_MODELfastembed model name for embeddings
DEFAULT_MODELDefault LLM alias: groq / smart / deep
NEXT_PUBLIC_API_URLFastAPI URL visible from the browser

8-Week Roadmap

Week 1 — Foundation

  • Project scaffold: FastAPI + Next.js + Docker Compose
  • Qdrant collection setup and ingest pipeline
  • PocketBase schema: sessions, messages, feedback

Week 2 — Knowledge Base

  • Ingest PR's bio, resume, project descriptions, blog posts
  • Tune chunking strategy (size, overlap, metadata)
  • Test retrieval quality with 20 benchmark queries

Week 3 — Chat Quality

  • Multi-turn conversation with session memory
  • System prompt iteration and persona tuning
  • Model routing (fast for simple, deep for complex)

Week 4 — UI Polish

  • Streaming responses (SSE)
  • Sidebar: real chat history from PocketBase
  • Mobile-responsive layout
  • Keyboard shortcuts (Enter to send, Shift+Enter newline)

Week 5 — Observability

  • Langfuse traces for every request
  • Cost tracking per session
  • Feedback loop: thumbs → PocketBase → Langfuse scores

Week 6 — Deployment

  • Dockerize API + UI
  • Deploy to Railway or Fly.io
  • Custom domain + HTTPS
  • Rate limiting and basic auth for public access

Week 7 — Enrichment

  • Auto-ingest from Notion (PR's notes)
  • Calendar availability endpoint ("Can I schedule a call?")
  • GitHub activity ingestion (recent repos, READMEs)

Week 8 — Launch

  • Public launch page with embed widget
  • Share link per conversation
  • Analytics dashboard (queries/day, top topics, avg cost)
  • README and documentation
Ver o original no GitHub

Bom para

Relacionados