Dylan Xu

Key Skills

Software

Data Annotation Tech

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Computer Code Programming

Text

Top Task Types

Action Recognition

Audio Recording

Computer Programming Coding

Evaluation Rating

Text Generation

Freelancer Overview

I’m an AI/LLM training-data specialist with cross-domain experience spanning tool-using assistants (Maps/Web/Yelp and other integrations), general programming copilots, and audio-specific QA. I label at the conversation and submission level using A/B (side-by-side) and Likert frameworks across axes such as factuality/grounding, tool-use correctness, instruction-following, safety/tone, clarity, and efficiency. My workflow includes pinpointing and fixing concrete errors (logic, compilation/runtime, type/IO, prompt adherence), writing step-by-step rationales, and adapting to evolving guidelines while maintaining consistent quality and throughput. What sets me apart is a strong intersection of audio post-production expertise (dialogue editing, restoration, deliverables/QC in Pro Tools, iZotope RX, Avid S4) and software depth (Python, Java, C++). I’ve supported quality control via second-pass audits and peer review/adjudication, documented edge cases, and contributed calibration notes to tighten rubrics and improve inter-rater consistency. Comfortable with rapid iteration and ambiguity, I bring clear communication, meticulous process habits, and practical engineering judgment to produce reliable, production-ready training data.

Labeling Experience

Speech Data QA

Data Annotation TechAudioAudio Recording

Performed quality checks on worker-recorded audio for speech datasets across varied devices and environments. Rated clips and flagged issues by axis (intelligibility, SNR/noise, clipping/distortion, loudness compliance, silence/pacing, channel balance/mono-stereo, prompt adherence/content safety, file/format validity, and metadata correctness). Verified technical specs and checked language/prompt compliance. Added targeted notes for re-record or acceptance with conditions, following evolving guideline updates and edge-case lists.

Performed quality checks on worker-recorded audio for speech datasets across varied devices and environments. Rated clips and flagged issues by axis (intelligibility, SNR/noise, clipping/distortion, loudness compliance, silence/pacing, channel balance/mono-stereo, prompt adherence/content safety, file/format validity, and metadata correctness). Verified technical specs and checked language/prompt compliance. Added targeted notes for re-record or acceptance with conditions, following evolving guideline updates and edge-case lists.

2025

Cross-Domain, Tool-Using LLM

Data Annotation TechComputer Code ProgrammingText GenerationEvaluation Rating

Evaluated a cross-domain, tool-using LLM that invoked Maps, Web Search, Yelp, and additional integrations for diverse real-world tasks. Performed conversation-level labeling by rating individual dialogues and pinpointing error types, then applied targeted edits to fix issues in line with evolving prompt guidelines. Supported quality control through second-pass audits and peer-review of other annotators’ work, adding adjudication notes and calibration feedback to uphold QA thresholds and guideline consistency.a Scale: multi-annotator, multi-month effort with thousands of items.

Evaluated a cross-domain, tool-using LLM that invoked Maps, Web Search, Yelp, and additional integrations for diverse real-world tasks. Performed conversation-level labeling by rating individual dialogues and pinpointing error types, then applied targeted edits to fix issues in line with evolving prompt guidelines. Supported quality control through second-pass audits and peer-review of other annotators’ work, adding adjudication notes and calibration feedback to uphold QA thresholds and guideline consistency.a Scale: multi-annotator, multi-month effort with thousands of items.

2025

General-Purpose Programming LLM

Data Annotation TechComputer Code ProgrammingEvaluation RatingComputer Programming Coding

Evaluated a general-purpose programming LLM across multiple languages for code generation, refactoring, debugging, and API-usage tasks. Performed conversation and submission-level labeling with side-by-side comparisons and Likert ratings across axes such as functional correctness, test conformance, clarity/style, complexity/efficiency, security/safety, and instruction-following. Identified and fixed specific errors (algorithmic edge cases, compilation/runtime issues, type/IO mismatches, inefficient patterns) via targeted edits in line with evolving prompt guidelines. Supported quality control through second-pass audits and peer review of other annotators’ work, contributing adjudication notes and calibration feedback to maintain rubric consistency and QA thresholds. (Where provided, verified outputs against unit tests, logs, or runner results.)

Evaluated a general-purpose programming LLM across multiple languages for code generation, refactoring, debugging, and API-usage tasks. Performed conversation and submission-level labeling with side-by-side comparisons and Likert ratings across axes such as functional correctness, test conformance, clarity/style, complexity/efficiency, security/safety, and instruction-following. Identified and fixed specific errors (algorithmic edge cases, compilation/runtime issues, type/IO mismatches, inefficient patterns) via targeted edits in line with evolving prompt guidelines. Supported quality control through second-pass audits and peer review of other annotators’ work, contributing adjudication notes and calibration feedback to maintain rubric consistency and QA thresholds. (Where provided, verified outputs against unit tests, logs, or runner results.)

2023

Education

Work History