antonmilev979/js-alert-ai-backend repository overview

⁠js-alert-ai-backend

FastAPI backend for the Joule AI Alert Monitoring system — an AI-assisted alert aggregation, enrichment, and analysis service designed for on-call engineers managing SAP Joule Studio landscapes.

⁠What it does

The backend ingests raw alert streams, enriches them with AI-driven root-cause analysis and dependency-graph reasoning, and exposes the results via REST and WebSocket APIs consumed by the real-time dashboard.

⁠Core pipeline

Upstream Alert Proxy → Ingestion → Enrichment → Broadcasting → Dashboard

1. Ingestion — polls the upstream alert FIFO, normalises and validates events, persists to PostgreSQL.
2. Enrichment — runs parallel AI workers (SAP AI Core / LLM) to classify alerts, map service dependencies, generate root-cause hypotheses, and build an AlertVerdict with confidence score.
3. Broadcasting — fans enriched events out to WebSocket subscribers, filtered per landscape.
4. Live Summary — continuously regenerates a plain-language situation report per landscape.

⁠Key features

• Real-time alert feed over WebSocket
• AI-generated root-cause hypotheses and blast-radius analysis
• Service dependency graph (topology)
• Natural-language Q&A about active alerts ("Ask Alerts"), with optional log file attachment
• Alert statistics, forecasting, and velocity analytics
• MS Teams webhook notifications
• MLflow tracing for LLM observability
• SAP Cloud Logging Service (CLS) integration for deep log correlation
• PostgreSQL in production; auto-fallback to SQLite for local development
• SAP XSUAA OAuth authentication (bypass flag available for local dev)

⁠API surface

⁠Production deployment

The image is designed to run on Kubernetes via the included Helm chart (deploy/helm/js-alert-ai/). In production it expects:

• A PostgreSQL instance (Kubernetes operator-managed)
• An SAP Cloud approuter sidecar handling OAuth (XSUAA)
• Valid SAP AI Core credentials for LLM enrichment

⁠Tech stack

• Python 3.11 / FastAPI / uvicorn
• SQLAlchemy + PostgreSQL / SQLite
• LangChain + SAP AI Core (gen_ai_hub)
• asyncio worker pool for pipeline stages

---