AI Observability Dashboard with GraphRAG
A multi-agent observability dashboard that fuses telemetry with knowledge graphs to speed up investigation workflows.
- langgraph
- graphrag
- observability
- memgraph
- python
Highlights
- Built a multi-agent workflow that correlates telemetry and graph context in a single timeline.
- Designed investigator-focused views that reduce mean time to root-cause.
- Integrated graph-backed retrieval to surface incident relationships and prior fixes.
Screenshots
What it is
A multi-agent observability dashboard that combines telemetry (metrics/logs/flow-style signals) with a knowledge-graph-driven retrieval layer (GraphRAG) to make investigations faster and more repeatable.
What I contributed
- Built the end-to-end demo: ingestion utilities, agent orchestration, and an interactive UI.
- Integrated a graph database for both adjacency queries and vector search so agents can retrieve relevant context.
- Focused on making the system “operator friendly”: clear prompts, bounded tool behavior, and explainable outputs.
Outcome / impact
- Turned raw observability data into a workflow-oriented network device analysis experience that answers questions like “Which devices are experiencing the most anomalies?” and “What is causing this device to shutdown frequently?”.
- Enabled the system to handle actions such as "Create an alert for these auth failures".
- Provided a practical pattern for combining LLM agents with structured graph context in an intuitive and actionable way.
Tech (high-level)
Python · LangGraph · GraphRAG · Memgraph · Panel UI · Vector indexing