ATLAS Framework
Automated Triaging and Log Analysis Solution
An AI root cause analysis tool for HPC clusters. Hand it a Jira ticket or an error log and LLM agents gather the evidence and write a root-cause report with remediation steps.
- Engineered agent context via tiered retrieval and 3-layer de-duplication so the LLM reads all evidence at once.
- Shipped semantic search with BGE-M3 embeddings and re-ranking across 3,291 Jira tickets.
- Launched the web UI and REST API with FastAPI and GitHub Copilot SDK, backed by 270+ pytest tests.
- Python
- Model Context Protocol
- FastAPI
- FAISS
- RAG
- GitHub Copilot SDK
Source stays inside HPE.
pipeline
- Jira ticket or error logPods restart in a loop after last night's upgrade.
- 01Health checks50+ automated checks
- 02Similarity searchFAISS over past tickets
- 03LLM root cause analysisagents with 15 MCP tools
- RCA reportin about 2 minutes
- Root cause
- The new chart renamed a config key that the service still reads at startup.
- Fix
- Add the new key to the values file and roll the deployment.
A simulated run on made-up incidents. The real pipeline takes about 2 minutes.
- ~80%
- less manual triage
- 15
- MCP tools
- ~2 min
- per RCA report








