Player not loading? Watch on YouTube
In this video sponsored by Oracle Developers, Rahul Wagh builds a local AI incident copilot that searches historical incidents, retrieves runbooks, and identifies service owners. The opening demo asks for an incident summary and returns records with service, severity, region, and summary fields.
The walkthrough explains how requests pass from Streamlit through FastAPI to a LangChain agent. Ollama runs the local LLM, identified in the source description as Llama 3.1 8B. LangChain connects it to Oracle Database 26ai vector search. The speaker introduces embeddings and nearest-neighbor retrieval with simple examples before configuring database users, tables, HNSW indexes, and B-tree filters.
Setup covers Docker Desktop, a Python virtual environment, dependency installation, and connectivity tests. The source description lists Docker Desktop 20.10+, Python 3.10+, and about 15 GB of free disk. The Ollama installation demo uses macOS and points to Windows and Linux instructions. The speaker suggests 16 GB RAM for the model.
The seed script loads 15 incidents, 15 runbooks, and 20 services. Four agent tools handle similar incidents, filtered searches, runbooks, and ownership. FastAPI runs on localhost port 8000; Streamlit uses 8501. The model runs locally after the initial software and model downloads.
