Skip to content

[Obs AI Assistant] Update evaluation script to consistently score all evaluation criteria #226983

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
SrdjanLL merged 2 commits into from
Jul 10, 2025
Merged

[Obs AI Assistant] Update evaluation script to consistently score all evaluation criteria #226983

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e08501c
Update evaluation script's system instructions and scoring tool param…
SrdjanLL Jul 8, 2025
afd9211
Use static system message
SrdjanLL Jul 8, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 13 additions & 1 deletion .../observability/plugins/observability_ai_assistant_app/scripts/evaluation/kibana_client.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -519,6 +519,8 @@ export class KibanaClient {
};
},
evaluate: async ({ messages, conversationId, errors }, criteria) => {
const criteriaCount = criteria.length;

const message = await chat('evaluate', {
connectorIdOverride: evaluationConnectorId,
systemMessage: `You are a critical assistant for evaluating conversations with the Elastic Observability AI Assistant,
Expand All @@ -527,7 +529,15 @@ export class KibanaClient {
Your goal is to verify whether a conversation between the user and the assistant matches the given criteria.

For each criterion, calculate a score. Explain your score, by describing what the assistant did right, and describing and quoting what the
assistant did wrong, where it could improve, and what the root cause was in case of a failure.`,
assistant did wrong, where it could improve, and what the root cause was in case of a failure.

### Scoring Contract

* You MUST call the function "scores" exactly once.
* The "criteria" array in the arguments MUST contain **one object for EVERY criterion**.
* If a criterion cannot be satisfied, still include it with \`"score": 0\` and a short \`"reasoning"\`.
* Do NOT omit, merge, or reorder indices.
* Do NOT place the scores in normal text; only in the "scores" function call.`,
messages: [
{
'@timestamp': new Date().toString(),
Expand Down Expand Up @@ -557,6 +567,8 @@ export class KibanaClient {
properties: {
criteria: {
type: 'array',
minLength: criteriaCount,
maxLength: criteriaCount,
items: {
type: 'object',
properties: {
Expand Down