⚡ AI Automation Command Center

Portfolio-grade AI automation control plane with LangGraph orchestration, asynchronous workers, human approval, evidence, audit history, observability, and a polished operations workspace.

AI Automation Command Center demonstrates how agentic systems become reliable business workflows rather than isolated LLM demos. The platform covers content generation, competitor intelligence, partner outreach, and KPI/leadership reporting.

The project now includes a substantially more product-oriented operations UI while preserving the workflow backend:

• Control-plane UX: System Pulse, System Health, workflow preview, approvals, audit trail, and recent execution views.
• Live workflow visualization: active runs expose high-level lifecycle states without fabricating per-node telemetry.
• One-click demo workflow: demo runs use the real /api/v1/runs path rather than simulated UI output.
• Execution details: run metadata, status, duration, output, and audit information are surfaced in the UI.
• Clear History: history can be cleared through a protected API path while active jobs are preserved.
• Reliable sidebar: native Streamlit sidebar collapse/reopen behavior is repaired for hosted deployments.
• Responsive theme-aware styling: the UI is designed as an operational control plane rather than a generic chatbot.

Most agent portfolios stop at prompt -> LLM -> answer. This project demonstrates the engineering layer around the model:

• LangGraph workflow orchestration
• FastAPI control plane with typed contracts
• Redis + RQ asynchronous job execution
• PostgreSQL durable run and audit storage
• Human-in-the-loop approval for external-impact workflows
• Evidence and validation instead of unsupported claims
• Scheduled automation through a dedicated scheduler service
• Prometheus + Grafana operational metrics
• OpenTelemetry + Jaeger distributed tracing
• Deterministic demo mode for portfolio demonstrations
• CI evaluation suite for workflow regression
• Docker Compose deployment

Streamlit Operations Console
 ↓
FastAPI Control Plane
 ↓
Redis Queue ─────────── PostgreSQL
 ↓ │
RQ Workers Audit / History
 ↓
LangGraph Workflows
 ├── Content Agent
 ├── Competitor Agent
 ├── Outreach Agent
 └── KPI Agent
 ↓
Validation + Evidence + Usage
 ↓
Completed / Failed / Awaiting Approval
Observability:
Prometheus → Grafana
OpenTelemetry → Jaeger

External-impact actions remain explicitly controlled. For outreach:

Candidate
 ↓
Qualification
 ↓
Draft
 ↓
approval_required = true
 ↓
Human reviewer
 ├── approved
 └── rejected

The approval decision is persisted in PostgreSQL and exposed in the audit trail. The UI intentionally treats autonomy and risk boundaries as separate concerns.

docker compose up --build -d
python scripts/smoke_test.py

UI: http://localhost:8501
API: http://localhost:8000
Swagger: http://localhost:8000/docs

docker compose run --rm api pytest -q
docker compose run --rm api ruff check .
docker compose run --rm api python -m compileall -q app ui tests scripts

AI operations control plane — agent orchestration, asynchronous execution, approval boundaries, observability, and auditability.