How vkuruganti/aiops works
Project type and purpose
• This is a microservices-based orchestration system for enterprise on-call triage.
• Two main areas:
• “agent” – a set of small AI/LLM-powered REST services (PDF processing, ping test, etc.)
• “oncall” – a Python orchestration pipeline that sequences alert ingestion, deduplication, summarization, question generation, action proposal, and ServiceNow ticket drafting, optionally offloading ML inference to Nvidia Triton.
File structure
• Top-level:
• README.MD
• agent/
• oncall/
• agent/
• app.py & ping_test.py – Flask entrypoints
• requirements.txt – Flask, PyMuPDF, OpenAI SDK
• pdf/ – sample documents for processing
• templates/index.html – minimal UI
• oncall/oncall-triage-orchestration/
• src/ – pure-Python modules by pipeline stage
• each subfolder is a self-contained component: ingestion, deduplication, summarization, clarifying_questions, next_actions, servicenow
• triton_client.py – Nvidia Triton gRPC client
• utils/helpers.py – shared code
• main.py – coordinator that wires modules in sequence
• .env & requirements.txt – environment variables and dependencies
README for additional insights
• agent folder = independent AI agents, designed to run as microservices
• oncall folder = orchestration logic, runs those agents or Triton for scalable inference
• Modularity and separation of concerns
Summary of steps for the principal software engineer
6) With protocols (HTTP, gRPC) and key library names (Flask, Triton client).