convez.ai
MCP ServerDados & AnalyticsAvançado
RajKumarK01

ASCENT — Sistema Multi-Agente de Preparação para Certificações

RajKumarK01/ASCENT

Preparação para certs com multi-agentes, planos adaptativos e insights por equipe

Ver no GitHub
InicialPluginScore · de 0 a 100

MCP Server que implementa um sistema de aprendizado corporativo multi-agente baseado em Microsoft Foundry, estruturando preparação para certificações profissionais. Orquestra especialistas em curadoria de conteúdo, planejamento adaptativo, engajamento contextual e avaliação com base em dados reais de trabalho e capacidade.

Instalação

$npx convez-ai RajKumarK01/ASCENT

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
42%
Último commit
há 5 meses
Versão
—
Issues abertas
0
Contribuidores
2

Casos de uso

  • Preparação estruturada para certificações Microsoft em grandes organizações
  • Planejamento de estudos adaptativo com base em carga de trabalho real
  • Avaliação de prontidão em nível de equipe sem exposição de dados pessoais

Veredito

Solução robusta e bem-arquitetada que resolve um problema real de corporações (escala de capacitação técnica). Acoplamento forte ao ecossistema Microsoft (Foundry, Fabric, Work IQ) limita reutilização; requer ambiente Azure produção para valor completo. Excelente referência de orquestração multi-agente.

README

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

ASCENT

Adaptive Skills & Certification ENablement for Teams A multi-agent enterprise learning system for the Reasoning Agents track (Battle #2), built on Microsoft Foundry + the Microsoft Agent Framework.

⚠️ All data and documents in this repository are SYNTHETIC and for demonstration only. No real personal, customer, employee, or proprietary data is used. Identifiers such as L-1001, EMP-001, and TEAM-A are fabricated.


What it does

A learner gives a certification goal. The system:

  1. Curates cited learning content for that goal and role (Foundry IQ).
  2. Builds a realistic, capacity-aware study plan (Fabric IQ semantic model).
  3. Schedules reminders around real work rhythm (Work IQ signals).
  4. Generates grounded, cited practice questions and scores readiness (Foundry IQ + Fabric IQ).
  5. Passes ready learners forward or loops back into preparation.
  6. Surfaces team-level readiness and risk to managers — without exposing PII.

Architecture

A top-level Orchestrator plans and routes to five specialist agents:

AgentRoleGrounding
OrchestratorPlan + route + assembleAgent Framework workflow
Learning Path CuratorCert → skills → cited contentFoundry IQ + Microsoft Learn MCP
Study Plan GeneratorContent → schedule + milestonesFabric IQ
Engagement AgentStudy windows around work loadWork IQ
Assessment AgentCited questions + readiness scoreFoundry IQ + Fabric IQ
Manager Insights AgentTeam readiness + risk (no PII)Fabric IQ + Work IQ

Reasoning patterns: Planner–Executor (Orchestrator), Critic/Verifier (Assessment + grounding check), self-reflection loop (pass/fail), role specialisation.


Quickstart (local — backend agents)

python -m venv .venv
# Windows
.venv\Scripts\activate
# macOS/Linux
source .venv/bin/activate

pip install -r requirements.txt
cp .env.example .env   # fill in your Foundry values

python -m src.orchestrator L-1001   # runs the full flow locally with stub fallbacks
python -m evals.run_evals           # runs the eval scorecard (23/23 expected)

The code runs locally with deterministic fallbacks even before Azure resources exist. Flip ASCENT_MODE=foundry in .env to switch to real Foundry + Azure AI Search.

Quickstart (local — full UI)

# Terminal 1 — FastAPI backend-for-frontend
pip install -r requirements.txt
python scripts/seed_users.py        # generates data/demo_users.json
uvicorn api.main:app --port 8000

# Terminal 2 — React frontend
cd ui
npm install
npm run dev                          # http://localhost:5173

Demo credentials

EmailPasswordRoleScope
emp.morgan@ascent.demodemo-pass-1EmployeeL-1001 (Cloud Eng, loops to READY)
emp.alex@ascent.demodemo-pass-2EmployeeL-1002 (DevOps, READY)
emp.casey@ascent.demodemo-pass-3EmployeeL-1004 (failing learner)
mgr.taylor@ascent.demodemo-mgrManagerTEAM-A
mgr.jordan@ascent.demodemo-mgr2ManagerAll teams

The chat panel on the Employee Dashboard accepts free-text queries such as:

  • "Am I ready to book the exam?"
  • "Build me a 6-week plan"
  • "Focus on Azure Functions"

Deploy (Hosted Agent on Foundry Agent Service)

azd auth login
azd provision   # creates Foundry project, model, App Insights, container registry
azd deploy      # builds container, pushes to ACR, publishes the hosted agent
azd ai agent show
azd ai agent invoke "Help L-1001 prepare for AZ-204"

Fallback: deploy the same container to Azure Container Apps if Hosted Agent quota is unavailable.


Repo layout

ascent/
├── README.md  .gitignore  .env.example
├── requirements.txt  azure.yaml  Dockerfile  agent.yaml
├── api/                # FastAPI BFF (auth, employee routes, manager routes, chat)
├── data/               # synthetic datasets (learners, work signals, semantic seed, demo users)
├── docs/               # synthetic knowledge docs (Foundry IQ source)
├── scripts/            # seed_users.py — generates demo_users.json
├── src/
│   ├── orchestrator.py  server.py  config.py
│   ├── agents/         # curator, study_planner, engagement, assessment, manager_insights
│   └── iq/             # foundry_iq, fabric_iq, work_iq
├── evals/              # testcases.json + run_evals.py (23 cases)
└── ui/                 # React + Vite + Tailwind SPA
    └── src/
        ├── pages/      # Login, EmployeeDashboard, StudyPlan, Assessment, ManagerDashboard
        └── components/ # ChatPanel, TraceTimeline, ProgressRing, CitationChip, Badge, Card
Ver o original no GitHub

Bom para

Relacionados