June 2025 Summaries

AI applications present unique challenges as their errors are not simply binary but can involve producing incorrect or misleading responses, necessitating a need for thorough observability into their reasoning processes. Observe has launched a public beta of LLM Observability to address this issue, offering tools like the LLM Explorer to enhance visibility into AI performance, costs, and behavior. This tool enables the investigation of AI response quality, cost optimization, and troubleshooting of AI infrastructure by providing insights into agent workflows, tracing reasoning chains, and examining prompt engineering. In a case study involving a bank's customer support chatbot, the tool helped identify a flaw in customer classification logic that led to incorrect recommendations, demonstrating the importance of understanding AI reasoning and infrastructure interdependencies. Additionally, the platform aids in real-time cost tracking and optimizing token usage, ensuring that AI applications remain cost-effective and reliable under real-world conditions.

Database performance issues, particularly the n+1 query pattern, are common challenges in microservice architectures. This pattern, where an application makes multiple inefficient database queries, can lead to significant performance bottlenecks. Observe APM offers a solution to diagnose and resolve these issues by efficiently tracing and identifying the root cause of latency spikes. In a real-world example, a developer investigating checkout service latency discovers that the cart service's `EmptyCart` endpoint is causing the problem due to multiple LREM operations in Redis, indicative of an n+1 issue. By using Observe APM, the developer can quickly identify and address the inefficiency by updating the logic to consolidate database queries, thus improving the checkout experience and reducing database load. The platform's reliance on OpenTelemetry ensures industry-standard instrumentation and future-proof observability strategies, making it a robust tool for tackling database performance bottlenecks.