convez.ai
MCP ServerBusca & PesquisaAvançado
navyforses

ALEKSANDRA_BRAIN v4 — Sistema MCP para Pesquisa Pediátrica

navyforses/ALEKSANDRA_BRAIN_v4

Cérebro IA 24/7 para diagnósticos pediátricos complexos com visualização 3D

Ver no GitHub
InicialPluginScore · de 0 a 100

Um servidor MCP de pesquisa médica contínua, especializado em HIE pediátrica, que integra 52 servidores MCP, 5 agentes autônomos e visualização 3D de neuroimagem. Arquitetura em 5 camadas (Percepção→Memória→Cognição→Visualização→Ação) que automatiza busca de papers, análise de relevância, descoberta de reposicionamento de fármacos e comunicação bidirecional com famílias.

Instalação

$npx convez-ai navyforses/ALEKSANDRA_BRAIN_v4

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 4 meses
Versão
—
Issues abertas
0
Contribuidores
4

Casos de uso

  • Pesquisa contínua e automática em fontes acadêmicas médicas
  • Análise automatizada de relevância de papers com scoring inteligente
  • Descoberta de reposicionamento de fármacos via grafos de conhecimento
  • Visualização 3D interativa de neuroimagem pediátrica (NIfTI, FreeSurfer)
  • Orquestração de múltiplos agentes de IA para raciocínio complexo

Veredito

Projeto extremamente ambicioso e especializado para pesquisa médica pediátrica contínua. Requer expertise significativa em MCP, orquestração de agentes, grafos de conhecimento e infraestrutura (Neo4j, Qdrant, Docker). Implementação de referência sólida para pipelines de descoberta científica autônomos, mas com curva de aprendizado muito íngreme.

README

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

ALEKSANDRA_BRAIN

Living research brain for pediatric Hypoxic-Ischemic Encephalopathy (HIE). An open-source AI system that never sleeps — continuously searching, analyzing, and discovering treatment possibilities for children with severe brain injury.

Status: Phase 2.5 closed; Phase 3 cognition entry ready; budget workflow fix in progress (2026-05-16) Version: v4.0 License: MIT


The story in one paragraph

On 28 August 2025, Aleksandra Jintcharadze was born in Tbilisi, Georgia. Severe HIE caused diffuse cystic encephalomalacia across her brain. Her brainstem survived intact. The healthcare system is not built for families facing rare or severe pediatric diagnoses — no single institution integrates research, treatment discovery, navigation, and visualization. ALEKSANDRA_BRAIN is the systemic integrator: a continuously operating AI research brain designed for the critical 0-to-2-year neuroplasticity window. Built for Aleksandra, designed to serve any family facing a complex pediatric diagnosis.

Full story → docs/Project_Description.md


Architecture in five layers

PERCEPTION  →  MEMORY  →  COGNITION  →  VISUALIZATION  →  ACTION
   eyes        memory       thought          sight          hands
LayerStack
PerceptionCrawl4AI, Firecrawl MCP, Browser Use, RAGFlow, n8n
MemoryNeo4j + Graphiti, Qdrant + fastembed, Supabase, mem0, LightRAG
CognitionCrewAI (5 agents), Claude Sonnet 4, DSPy, Adaptive GoT
VisualizationNiiVue, React Three Fiber, fork of FreeSurfer/freebrowse
ActionTelegram 2-way, Gmail, Notion, Google Calendar

Full architecture → docs/v4_MASTER.md


Five autonomous agents

#AgentCadenceRole
1SpiderEvery 6 hoursHunts new papers across 15+ academic sources
2AnalyzerPer paperScores relevance, extracts entities, stores in KG
3HypothesisWeeklyCross-disease pattern finder
4RepurposingMonthlyOpen Targets → DrugBank → PubChem → L1000
5CommunicatorContinuousFamily liaison via Telegram + email

Source → agents/


Five world-firsts (as of May 2026)

  1. First MCP server for neuroimaging (NIfTI, FreeSurfer, atlas operations).
  2. Patient-specific 3D brain with educational-quality visuals (NiiVue + R3F).
  3. Continuously running drug repurposing pipeline for neonatal HIE.
  4. Temporal knowledge graph for medical evidence with automatic confidence decay.
  5. Two-way evidence gating between AI and family via Telegram.

Quick start

Prerequisites

  • Python 3.11+
  • Node 20+
  • Docker Desktop (for Neo4j + Qdrant)
  • Accounts: Supabase, Neo4j AuraDB, Cloudflare, Railway, Anthropic

1. Clone and configure

git clone https://github.com/jincharadzeshako/ALEKSANDRA_BRAIN.git
cd ALEKSANDRA_BRAIN
cp .env.example .env
# edit .env with real credentials

2. Install Claude Code productivity plugins (one-time)

# Windows PowerShell — run from repo root
./setup-claude-code.ps1

This installs Claude Mem, Caveman (Full mode), GSD, Graphify, and Code Review Graph. Full guide → docs/CLAUDE_CODE_SETUP.md.

3. Local infrastructure

docker-compose up -d        # Neo4j on :7474, Qdrant on :6333
pip install -r requirements.txt

4. Current verification

.venv/Scripts/python.exe -X utf8 -m scripts.verify_phase1
.venv/Scripts/python.exe -X utf8 -m scripts.verify_phase2 --gate all
.venv/Scripts/python.exe -X utf8 -m scripts.verify_phase2_5 --gate all
.venv/Scripts/python.exe -X utf8 scripts/test_crew.py

On Windows, use the project virtualenv and -X utf8; bare python can miss dependencies and CP1252 console output can break verifier tables.

5. Run the brain shell

python -m mcp.hello_brain      # MCP server

Current progress

Historical Phase 0 materials remain available: docs/PHASE_0_PLAN.md and docs/PHASE_0_HANDOUT.md.

Archived audit snapshots remain available under docs/archive/: PHASE_2_LIVE_AUDIT.md and ACTIVITY_DIAGNOSTIC_PLAN.md.

Budget caveat: Phase 2.5A spend instrumentation passes, but the deployed n8n daily-budget-gate JSON-body expression bug is still being fixed. Treat the code-side check_daily_budget() guard as active; treat workflow-written budget_lock rows as pending confirmation until that fix lands.


Principles

  • Unknown potential, not limited outcomes. Never write a pessimistic prognosis. MRI structure is not a function ceiling.
  • 0-2 years is the plasticity window. Every week matters.
  • Never invent a fact. No source found → "not found".
  • MRI stays client-side. Patient imaging never leaves the family's browser.
  • A real doctor signs every decision. This is an information system, not a clinician.

Cost envelope

  • MVP: $20-30 / month (Supabase free, Neo4j Aura free, Qdrant Docker, n8n Railway hobby, Vercel hobby, Claude API metered).
  • Full: ~$120 / month (paid tiers as data grows).

License

MIT. Built with love and urgency. For Aleksandra and every family the system was never built for.

Ver o original no GitHub

Bom para

Relacionados