Agents, LLM Evals, OpenTelemetry – Lessons from Building Arize Phoenix in 2024

The team behind Arize Phoenix, an open-source LLM observability tool, reflects on a year of rapid growth and industry changes. Over the past year, the AI landscape has evolved significantly, with agents becoming standard, LLM evaluation turning into a product necessity, and OpenTelemetry establishing itself as the observability benchmark. During this time, Arize Phoenix's monthly downloads skyrocketed from 20,000 to 2.5 million. The team gained insights into adapting OpenTelemetry for LLMs, the need for improved evaluation frameworks for agents, and the integration of synthetic and non-synthetic data for evaluations. Despite challenges, these efforts contributed to advancing the field.

The community's support has been instrumental in driving development, leading to significant launches like Prompt Playground and Sessions in 2024. The team expresses gratitude to users who engaged with Arize Phoenix by providing feedback and participating in meetups. They remain committed to further innovation and development, with more projects on the horizon. For a detailed reflection, readers are encouraged to visit their blog.

Key takeaways

Arize Phoenix, an open-source LLM observability tool, saw rapid growth from 20k to 2.5M monthly downloads in a year.
The team adapted OpenTelemetry for LLM observability and developed better evaluation frameworks for agents.
They effectively combined synthetic and non-synthetic data for evaluations, contributing to advancements in the AI industry.
Major launches in 2024 included Prompt Playground and Sessions, with more developments planned for the future.

Agents, LLM Evals, OpenTelemetry – Lessons from Building Arize Phoenix in 2024

Key takeaways

Discussion (0)