Designing Rootify as a Practical RAG System

Rootify is the kind of project I want this portfolio to explain well: not just what was built, but why the shape of the system matters.

The useful version of a log-analysis assistant is not a chatbot pasted on top of Elasticsearch. It needs a retrieval strategy, source evidence, transaction boundaries, and a way to explain uncertainty.

The rough architecture I would present:

• Ingest logs from multiple services and normalize common fields.
• Group events by transaction ID, user journey, or request correlation ID.
• Retrieve exact matches through Elasticsearch and related historical context through embeddings.
• Ask the LLM to produce a hypothesis with citations back to the retrieved evidence.
• Render separate views for engineering RCA and business impact.

The interesting engineering constraint is trust. If an AI system says "service X caused the failure," the interface should make it cheap to inspect the trace that led there.

That is the design direction I like: AI that speeds up investigation while keeping the engineer in control.