Overview
An AI agent system designed to analyze incidents through multi-step reasoning workflows. Built with LangGraph and tested with local LLM setups via Ollama for fast experimentation.
Problem Statement
Incident analysis often requires combining context, logs, and assumptions into a structured investigation. The goal was to prototype an agentic workflow that can decompose a problem, reason step-by-step, and produce actionable summaries.
System Architecture
- Orchestration: LangGraph multi-step workflows
- Tools: retrieval / parsing / structured outputs (prototype)
- LLM Runtime: local inference via Ollama
Technologies Used
Python
LangChain
LangGraph
Ollama
LLM Agents
What I Learned
- Designing agent workflows with explicit steps and guardrails
- Rapid iteration with local LLMs
- Structuring outputs for clarity and debugging