@arizeai/phoenix-evals/src/llm/createUserFrictionEvaluator.ts

A library for running evaluations for AI use cases

import { USER_FRICTION_CLASSIFICATION_EVALUATOR_CONFIG } from "../__generated__/default_templates";
import type { CreateClassificationEvaluatorArgs } from "../types/evals";
import type { ClassificationEvaluator } from "./ClassificationEvaluator";
import { createClassificationEvaluator } from "./createClassificationEvaluator";

export interface UserFrictionEvaluatorArgs<
  RecordType extends Record<string, unknown> = UserFrictionEvaluationRecord,
> extends Omit<
  CreateClassificationEvaluatorArgs<RecordType>,
  "promptTemplate" | "choices" | "optimizationDirection" | "name"
> {
  optimizationDirection?: CreateClassificationEvaluatorArgs<RecordType>["optimizationDirection"];
  name?: CreateClassificationEvaluatorArgs<RecordType>["name"];
  choices?: CreateClassificationEvaluatorArgs<RecordType>["choices"];
  promptTemplate?: CreateClassificationEvaluatorArgs<RecordType>["promptTemplate"];
}

/** A conversation and the latest user message to evaluate for expressed friction. */
export interface UserFrictionEvaluationRecord {
  /**
   * Human-readable conversation history before the target user message.
   */
  conversation: string;
  /**
   * The latest user message to classify for expressed friction.
   */
  userMessage: string;
  [key: string]: unknown;
}

/**
 * Creates a user friction evaluator function.
 *
 * This function returns an evaluator that detects expressed user friction
 * with an assistant's preceding behavior, such as corrections, retries,
 * frustration, and challenges to unrequested or unexplained actions.
 *
 * @param args - The arguments for creating the user friction evaluator.
 * @param args.model - The model to use for classification.
 * @param args.choices - The possible classification choices (defaults to friction/no_friction).
 * @param args.promptTemplate - The prompt template to use (defaults to USER_FRICTION_CLASSIFICATION_EVALUATOR_CONFIG.template).
 * @param args.telemetry - The telemetry to use for the evaluator.
 *
 * @returns An evaluator function that takes a {@link UserFrictionEvaluationRecord} and returns a classification result
 * indicating whether the latest user message expresses friction or no friction.
 *
 * @example
 * ```ts
 * const evaluator = createUserFrictionEvaluator({ model: openai("gpt-4o-mini") });
 * const result = await evaluator.evaluate({
 *   conversation: "User: Show recent orders.\nAssistant: Here are last month's orders.",
 *   userMessage: "No, I asked for this week.",
 * });
 * console.log(result.label); // "friction"
 * ```
 */
export function createUserFrictionEvaluator<
  RecordType extends Record<string, unknown> = UserFrictionEvaluationRecord,
>(
  args: UserFrictionEvaluatorArgs<RecordType>
): ClassificationEvaluator<RecordType> {
  const {
    choices = USER_FRICTION_CLASSIFICATION_EVALUATOR_CONFIG.choices,
    promptTemplate = USER_FRICTION_CLASSIFICATION_EVALUATOR_CONFIG.template,
    optimizationDirection = USER_FRICTION_CLASSIFICATION_EVALUATOR_CONFIG.optimizationDirection,
    name = USER_FRICTION_CLASSIFICATION_EVALUATOR_CONFIG.name,
    ...rest
  } = args;
  return createClassificationEvaluator<RecordType>({
    ...rest,
    promptTemplate,
    choices,
    optimizationDirection,
    name,
  });
}