feat: add multimodal online-evaluation guide

Logs multimodal (text + image) traces and creates the online LLM-as-judge
rule that scores them — both in the Opik UI (step-by-step) and with the SDK
(rest_client.automation_rule_evaluators). The image is supplied as an
image_url content part, not a text variable; the notebook uses env-var-only
credentials with a DRY_RUN fallback. Verified against a live workspace:
traces log and the SDK rule round-trips.

Includes a README (4 sections + Colab badge) and root/guides index entries.

Author: LeoRoccoBreedt

README.md

@@ -33,6 +33,7 @@ Task-oriented examples for specific Opik workflows and patterns.
 | | Description |
 |---|---|
 | [guides/annotation_queues_with_context](guides/annotation_queues_with_context/) | Structure RAG traces for Opik annotation queues — clean answer in output, context in metadata, full detail in child spans |
+| [guides/multimodal_online_evaluation](guides/multimodal_online_evaluation/) | Online LLM-as-judge evaluation over multimodal (text + image) traces, created both in the UI and via the SDK |
 | [guides/tracing_finetuned_models](guides/tracing_finetuned_models/) | Fine-tune a model, register it to the CometML Model Registry, then fetch and trace inference in Opik |
 
 ## Use Cases

guides/README.md

@@ -5,6 +5,7 @@ Task-oriented examples for doing something specific with Opik — combining Opik
 | Guide | Description |
 |---|---|
 | [annotation_queues_with_context/](./annotation_queues_with_context/) | Structure RAG traces for Opik annotation queues — clean answer in output, context in metadata, full detail in child spans |
+| [multimodal_online_evaluation/](./multimodal_online_evaluation/) | Run an online LLM-as-judge eval over multimodal (text + image) traces — create the rule in the UI and with the SDK |
 | [tracing_finetuned_models/](./tracing_finetuned_models/) | Fine-tune a model, register it to the CometML Model Registry, then fetch and trace inference in Opik |
 
 [Contribute one](../CONTRIBUTING.md).

guides/multimodal_online_evaluation/.gitignore

@@ -0,0 +1,3 @@
+.venv/
+uv.lock
+.ipynb_checkpoints/

guides/multimodal_online_evaluation/README.md

@@ -0,0 +1,36 @@
+# Multimodal Online Evaluation
+
+Run an online LLM-as-judge evaluation over multimodal (text + image) traces in Opik, and create the scoring rule two ways — in the UI and with the SDK.
+
+## What this does
+
+Logs traces that carry an image plus text, then shows how to score them with an online LLM-as-judge rule. It documents the common footguns — add image variables with the **"Images +"** button (not by hand-typing `{{image_output}}`), use a vision-capable model, and remember that score definitions are independent of variable mapping — and creates the identical rule both in the UI and programmatically with the SDK.
+
+## Prerequisites
+
+```bash
+uv sync    # or: pip install "opik>=2.0.74"
+```
+
+| Variable | Description |
+|---|---|
+| `OPIK_API_KEY` | Opik API key |
+| `OPIK_WORKSPACE` | Opik workspace name |
+| `OPIK_URL_OVERRIDE` | Optional; defaults to `https://www.comet.com/opik/api` |
+
+No LLM provider key is needed here — the judge runs inside Opik's online-evaluation rule, not in this notebook. Without credentials the notebook runs in dry-run and prints the payloads instead of sending them.
+
+## Running it
+
+```bash
+uv sync
+uv run --with jupyter jupyter lab
+```
+
+Then open `multimodal_online_evaluation.ipynb`.
+
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/comet-ml/opik-examples/blob/main/guides/multimodal_online_evaluation/multimodal_online_evaluation.ipynb)
+
+## How it works
+
+The notebook logs two multimodal traces (one image as a public URL, one as a base64 data URI), then walks through creating the online-evaluation rule. The UI path is documented step by step with a copy-paste judge prompt; the SDK path creates the same rule via `client.rest_client.automation_rule_evaluators`, with the image supplied as an `image_url` content part rather than a text variable. The final section covers running the rule and reviewing the feedback scores.