Efficiently tag LLM interactions with PostHog for quick analysis #tagging

PostHog-LLM is a tool that enhances PostHog, an analytics tool for tracking user interactions on the web, by adding the capability to visualize dialogs during Product Insights. It simplifies data labeling during data ingestion by leveraging fast ML models and Apps feature. Three plugins have been developed: a toxicity classifier, an emotions classifier, and an event flow classifier. The event flow plugin tracks common patterns in user and agent dialogs. The blog post explains how to upload user-agent data into PostHog-LLM and install the event flow plugin. The installation process involves deploying a fully containerized Posthog-LLM hobby instance locally using Docker and installing the PostHog Python SDK. Once set up, the flow detection plugin can be installed by pasting a specific link. Data can be sent from the ShareGPT dataset, and a script is provided to send the data using exported variables. The post also explains how to visualize and analyze the data in Product Analytics, providing insights from the sample data. Overall, PostHog-LLM offers a comprehensive solution for tracking and analyzing user interactions on the web.

Source link

Source link: https://medium.com/@afilipe362/quickly-load-and-label-your-llm-interactions-with-posthog-llm-8a748e7b5643?source=rss——llm-5