Chatting with logs: fine-tuning LLMs for LogQL

Description

Can LLMs write your log queries for you? 🤔 At Rootconf Mini 2024, Siddharth Balyan presents an exploratory deep dive into how LLMs can be fine-tuned to understand LogQL, Grafana’s log query language — pushing the boundaries of AI-assisted observability. This talk shares the team’s research on creating the first-ever text-to-LogQL dataset, fine-tuning LLMs to translate natural language into log queries, and evaluating their effectiveness across different approaches. 💡 What you’ll learn: 1. Why existing log query interfaces are painfully unintuitive — and how AI can help. 2. How to build a dataset from scratch for a domain-specific language like LogQL. 3. Insights from fine-tuning and evaluating LLMs for structured query generation. 4. Early steps toward a semantic search engine for observability data. 🧠 Who should watch: * SREs, infra, and platform engineers working with logs and monitoring tools. * AI and MLOps practitioners exploring domain-specific fine-tuning. * Anyone who’s ever written a LogQL, DatadogQL, or SPL query and thought: “There has to be a better way.” 👩‍💻 Why this matters: This talk marks an exciting step toward making observability more intelligent — using LLMs not just to summarize logs, but to generate and reason with them. 📺 Watch the full talk from Rootconf Mini 2024 and explore more deep dives into the intersection of AI, systems, and infrastructure. 👉 Subscribe to Rootconf for more talks, workshops, and community events: hasgeek.com/rootconf#memberships