Moda blog

Latest

The new hype of classifiers is LARP

Moda uses in-house classifiers to help find agent mistakes. We tested twelve decision systems and versions as possible replacements.

· 6 min read

Recent

Guides

Langfuse vs LangSmith: tracing, evals, pricing, and self-hosting compared

Langfuse vs LangSmith on their own docs: tracing, evals, prompts, OpenTelemetry, self-hosting, framework lock-in, and pricing with a worked example.

LLM as a judge: how to write, calibrate and run judges on agent traces

How LLM as a judge works: judge prompt templates, position and verbosity bias, calibration against human labels, cost, and judges on production traces.

Hallucinations in AI agents: how to detect and prevent them

How to detect and prevent hallucinations in AI agents: fabricated tool calls, wrong arguments, invented facts, false success claims, and context rot.

LLM observability tools compared: a buyer’s guide for AI agents

Compare Langfuse, LangSmith, Braintrust, Arize Phoenix, Helicone, Datadog, OpenLLMetry and Moda: what each LLM observability tool is good at, and what it costs.

Tool call errors in AI agents: how to detect, classify and reduce them

Why AI agent tool calls fail, how to detect and classify tool call errors in production, which ones to retry, and how to reduce them with schemas and evals.

Detecting user frustration and sentiment in AI agent traces

Detect user frustration and sentiment in AI agent traces, measure it per turn, task or trace, and connect it to NPS-style signals and retention.

Silent failures in AI agents: how to detect and fix them

Silent failures in AI agents raise no alert. Learn to detect silent agent failures with worked examples, tool-result checks, and outcome verification.