Evidently AI Blog

We put together 7 examples of how top companies like Asana and GitHub run LLM evaluations. They share how they approach the task, what methods and metrics they use, what they test for, and their learnings along the way.

LLM safety benchmarks help to ensure models are robust and reliable. In this blog, we highlight 10 key safety and bias LLM benchmarks that help assess and improve LLM reliability.

Evidently open-source now has more tools for evaluating RAG. You can score context relevance, evaluate generation quality, and use different LLMs as evaluators.

RAG helps make LLM systems more accurate and reliable. We compiled 10 real-world examples of how companies use RAG to improve customer experience, automate routine tasks, and improve productivity.

The Evidently API is evolving — and it’s getting better! We are updating the open-source Evidently API to make it simpler, more flexible, and easier to use. Explore the new features.

In this guide, we’ll discuss key AI regulations, such as the EU AI Act and the Blueprint for AI Bill of Rights, and explain what they mean for teams building AI-powered products.

Real-world examples of LLM hallucinations and other failures that can occur in LLM-powered products in the wild, such as prompt injection and out-of-scope usage scenarios.

From being biased to making things up, there are numerous instances where we’ve seen AI going wrong. In this post, we’ll explore ten notable AI failures when the technology didn’t perform as expected.

This tutorial shows how to create, tune, and use LLM judges. We'll make a toy dataset and assess correctness and verbosity. You can apply the same workflow for other criteria.

We are launching Evidently Cloud, a collaborative AI observability platform built for teams developing products with LLMs. It includes tracing, datasets, evals, and a no-code workflow. Check it out!

How do companies use LLMs in production? We compiled 45 real-world LLM applications from companies that share their learnings from building LLM systems.

In this tutorial, we introduce the end-to-end workflow of LLM regression testing. You will learn how to run regression testing as a process and build a dashboard to monitor the test results.

In this tutorial, you will learn how to systematically check the quality of LLM outputs. You will work with issues like changes in answer content, length, or tone, and see which methods can detect them.

Our CTO Emeli Dral was an instructor for the ML Monitoring module of MLOps Zoomcamp 2024, a free MLOps course. We summarized the ML monitoring course notes and linked to all the practical videos.

We put together the most interesting conferences on AI, Machine Learning, and Data Science in 2025. And the best part? Some of them are free to attend or publish the content after the event.

Evidently open-source Python library now supports evaluations for LLM-based applications, including RAGs and chatbots. You can compare, test, and monitor your LLM system quality from development to production.

Did you miss some of the latest updates at Evidently open-source Python library? We summed up a few features we shipped recently in one blog.

In this tutorial, you will learn how to run batch ML model inference and deploy a model monitoring dashboard for production ML models using open-source tools.

Looking for MLOps courses to attend in 2023? We put together five great online MLOps courses for data scientists and ML engineers. They are free to join or publish their content for everyone to access without a fee.

How do different companies start and scale their MLOps practices? In this blog, we share a story of how DeepL monitors ML models in production using open-source tools.

A beginner-friendly MLOps tutorial on how to evaluate ML data quality, data drift, model performance in production, and track them all over time using open-source tools.

Evidently 0.4 is here! Meet a new feature: Evidently user interface for ML monitoring. You can now track how your ML models perform over time and bring all your checks to one central dashboard.

How do you monitor unstructured text data? In this code tutorial, we’ll explore how to track interpretable text descriptors that help assign specific properties to every text.

In this code tutorial, you will learn how to create interactive visual ML model cards to document your models and data using Evidently, an open-source Python library.

In this code tutorial, you will learn how to set up an ML monitoring system for models deployed with FastAPI. This is a complete deployment blueprint for ML serving and monitoring using open-source tools.

Monitoring embedding drift is relevant for the production use of LLM and NLP models. We ran experiments to compare 5 drift detection methods. Here is what we found.

In this blog, we recap the Ask-Me-Anything session with Lina Weichbrodt. We chatted about ML monitoring and debugging, adopting LLMs, and the challenges of being a freelance ML engineer.

In this code tutorial, you will learn how to run batch ML model inference, collect data and ML model quality monitoring metrics, and visualize them on a live dashboard.

In this tutorial, you will learn how to implement Evidently checks as part of an ML pipeline and send email notifications based on a defined condition.

How do different companies start and scale their MLOps practices? In this blog, we share a story of how Wayflyer creates ML model cards using open-source tools.

In this tutorial, you will learn how to create a data quality and ML model monitoring dashboard using the two open-source libraries: Evidently and Streamlit.

In this blog, we recap the Ask-Me-Anything session with Stefan Krawczyk. We chatted about how to build an ML platform and what data science teams do wrong about ML dataflows.

How to approach building an internal ML platform if you’re not Google? We put together stories from 10 companies that shared their platforms’ design and learnings along the way.

In this tutorial, we will explore issues affecting the performance of NLP models in production, imitate them on an example toy dataset, and show how to monitor and debug them.

In this blog, we recap the Ask-Me-Anything session with Neal Lathia. We chatted about career paths of an ML Engineer, building and expanding ML teams, Monzo’s ML stack, and 2023 ML trends.

Meet the new feature: data quality monitoring and drift detection for text data! You can now use the Evidently open-source Python library to evaluate, test, and monitor text data.

Want to know how companies with top engineering teams do machine learning? We put together a list of the best machine learning blogs from companies that share specific ML use cases, lessons learned from building ML platforms, and insights into the tech they use.

In this blog, we recap the Ask-Me-Anything session with Ben Wilson. We chatted about AutoML use cases, deploying ML models to production, and how one can learn about ML engineering.

We are thrilled to announce our latest and largest release: Evidently 0.2. In this blog, we give an overview of what Evidently is now.

In this series of blogs, we are showcasing specific features of the Evidently open-source ML monitoring library. Meet NoTargetPerformance test preset!

In this blog, we recap the Ask-Me-Anything session with Rick Lamers, where we chatted about the evolution of orchestration tools, their place within the MLOps landscape, the future of data pipelines, and building an open-source project amidst the economic crisis.

Now that Hacktoberfest 2022 is over, it’s time to celebrate our contributors, look back at what we’ve achieved together, and share what we’ve learned during this month of giving back to the community through contributing to open source.

In Evidently v0.1.59, we moved the existing dashboard functionality to the new API. Here is a quick guide on migrating from the old to the new API. In short, it is very, very easy.

Imagine you have a machine learning model in production, and some features are very volatile. Their distributions are not stable. What should you do with those? Should you just throw them away?

In this blog, we recap the Ask-Me-Anything session with Jacopo Tagliabue, where we chatted about ML at a reasonable scale, testing RecSys, MLOps anti-patterns, what’s hot in DataOps, fundamentals in MLOps, and more.

In this blog, we recap the Ask-Me-Anything session with Bozhao Yu and Sean Sheng, where we chatted about why deploying a model is hard, beginner mistakes and how to avoid them, the challenges of building an open-source product, and BentoML’s roadmap.

There is an overwhelming set of potential metrics to monitor. In this blog, we'll try to introduce a reasonable hierarchy.

In this blog, we recap Ask-Me-Anything session with Doris Xin, that covered the roles of Data Scientists and Data Engineers in an ML cycle, automation, MLOps tooling, bridging the gap between development and production, and more.

We recap Ask-Me-Anything session with Fabiana Clemente, which covered synthetic data, its quality, beginner mistakes in data generation, the data-centric approach, and how well companies are doing in getting there.

When one mentions "ML monitoring," this can mean many things. Are you tracking service latency? Model accuracy? Data quality? This blog organizes everything one can look at in a single framework.

Meet the new feature in the Evidently open-source Python library! You can easily integrate data and model checks into your ML pipeline with a clear success/fail result. It comes with presets and defaults to make the configuration painless.

We ran an experiment to help build an intuition on how popular drift detection methods behave. In this blog, we share the key takeaways and the code to run the tests on your data.

In this blog we recap Ask-Me-Anything session with Matt Squire, that covered MLOps maturity and future, how MLOps fits in data-centric AI, and why open-source wins.

Our CTO Emeli Dral gave a tutorial on how to use Evidently at the Stanford Winter 2022 course CS 329S on Machine Learning System design. Here is the written version of the tutorial and a code example.

In this blog we recap Ask-Me-Anything session with Hamza Tahir, that covered MLOps trends and tools, the future of real-time ML, and building an open-source startup.

In this blog we recap the second Evidently Community Call that covers the recent feature updates in our open-source ML monitoring tool.

In this blog we recap Ask-Me-Anything session with Alexey Grigorev, that covered all things production machine learning, from tools to workflow, and even a bit on community building.

Data and prediction drift often need contextual interpretation. In this blog, we walk you through possible scenarios for when you detect these types of drift together or independently.

Meet the new Data Quality report in the Evidently open-source Python library! You can use it to explore your dataset and track feature statistics and behavior changes.

We are building an open-source tool to evaluate, monitor, and debug machine learning models in production. Here is a look back at what has happened at Evidently AI in 2021.

Now, you can easily customize the pre-built Evidently reports to add your metrics, statistical tests or change the look of the dashboards with a bit of Python code.

Even if you can calculate the model quality metric, monitoring data and prediction drift can be often useful. Let’s consider a few examples when it makes sense to track the distributions of the model inputs and outputs.

What can you do once you detect data drift for a production ML model? Here is an introductory overview of the possible steps.

Now, you can use Evidently to display dashboards not only in Jupyter notebook but also in Colab, Kaggle, and Deepnote.

When monitoring ML models in production, we can apply different techniques. Data drift and outlier detection are among those. What is the difference? Here is a visual explanation.

You can use Evidently together with Prometheus and Grafana to set up live monitoring dashboards. We created an integration example for Data Drift monitoring. You can easily configure it to use with your existing ML service.

You can look at historical drift in data to understand how your data changes and choose the monitoring thresholds. Here is an example with Evidently, Plotly, Mlflow, and some Python code.

Is it time to retrain your machine learning model? Even though data science is all about… data, the answer to this question is surprisingly often based on a gut feeling. Can we do better?

Now, you can use Evidently to generate JSON profiles. It makes it easy to send metrics and test results elsewhere.

Can you train a machine learning model to predict your model’s mistakes? Nothing stops you from trying. But chances are, you are better off without it.

There is more to performance than accuracy. In this tutorial, we explore how to evaluate the behavior of a classification model before production use.

You can now use Evidently to analyze the performance of classification models in production and explore the errors they make.

What can go wrong with ML model in production? Here is a story of how we trained a model, simulated deployment, and analyzed its gradual decay.

You can now use Evidently to analyze the performance of production ML models and explore their weak spots.

Our second report is released! Now, you can use Evidently to explore the changes in your target function and model predictions.

We are excited to announce our first release. You can now use Evidently open-source python package to estimate and explore data drift for machine learning models.

No model lasts forever. While the data quality can be fine, the model itself can start degrading. A few terms are used in this context. Let’s dive in.

A bunch of things can go wrong with the data that goes into a machine learning model. Our goal is to catch them on time.

Garbage in is garbage out. Input data is a crucial component of a machine learning system. Whether or not you have immediate feedback, your monitoring starts here.

Who should care about machine learning monitoring? The short answer: everyone who cares about the model's impact on business.

Congratulations! Your machine learning model is now live. Many models never make it that far. Some claim, as much as 87% are never deployed.