Khadija Sana

Freelance AI evaluator with 3+ years verifying model and agent outputs against defined quality standards, including conversational agent behavior, multi-step task correctness, transcription, translation, and TTS data, across Appen/CrowdGen, Mindrift, and Toloka. Experienced writing and applying structured rubrics to judge task success, calibrating consistently across large volumes of ambiguous cases, and documenting reasoning clearly enough for other raters to follow. Certified AI Language Evaluator & Annotator (micro1).

Native

Urdu and Punjabi speaker with C1 English, adding a linguistic and cultural accuracy layer to multilingual AI training and evaluation data.

LLM Hazard Response Evaluator - Toloka - Freelance

(2026-01)

• Review model responses to safety-sensitive prompts against defined hazard and policy criteria, judging whether the response met the required standard and documenting the reasoning behind each judgment.
• Apply consistent evaluation criteria across high volumes of ambiguous, edge-case scenarios, functioning as the verification layer that determines whether a response counts as correct.

AI Agent Evaluator - Mindrift - Freelance

(2026-07)

• Interact with shopping and conversational agents using realistic multi-step queries, then score full task outcomes against structured rubrics covering safety, relevance, and agent behavior, with independent written input beyond scored fields.
• Assess first-person (egocentric) video data for quality, relevance, and usability as AI training data.

AI Language & Audio Evaluator - Appen/CrowdGen - Part-time

(2026-06)

• Urdu–English transcription project: transcribe and quality-check spoken Urdu against strict accuracy and style guidelines, passing calibration at 100% accuracy (Poliwag Audio Transcription QA, ur-PK).
• Punjabi translation project: review and quality-check Punjabi-language translation output for accuracy and cultural appropriateness.
• TTS Evaluation (Project A12852): review text-to-speech output for prosodic accuracy, clarity, and linguistic appropriateness.
• Document quality issues and edge cases in detailed, evidence-based reports supporting rater calibration and QA process improvement.

BS Zoology - Zoology - University of Sargodha, Pakistan (2015-01 - 2019-01)