Wang Q.

Key Skills

Software

Don't disclose

Dataloop

Top Subject Matter

Dialogue QA and instruction-following data for foundation model training

E-commerce product image attribute labeling for generative modeling

Agent reasoning-chain annotation for travel O2O plugin/function calling

Top Data Types

Text

Image

Document

Top Task Types

Prompt + Response Writing (SFT)

Entity (NER) Classification

Function Calling

Fine-tuning

Segmentation

Data Collection

Freelancer Overview

Data Annotator — Single-Turn Dialogue Data Annotation. Core strengths include Don't disclose, Internal, and Proprietary Tooling. AI-training focus includes data types such as Text, Image, and Computer Code and labeling workflows including Prompt + Response Writing (SFT), Entity (NER) Classification, and Function Calling.

Labeling Experience

AI Trainer — Competitor Analysis

Don't discloseText

Performed competitor analysis by running horizontal comparative tests of mainstream architectural AI products across multiple evaluation dimensions. Used comparative data to guide subsequent model fine-tuning direction and sampled response quality. • Compared products across professionalism, logic rigor, and multimodal granularity • Assessed response accuracy, instruction following, and legal compliance • Guided next-stage fine-tuning through comparative data analysis • Sampled response quality to inform optimization priorities

2024 - 2026

AI Trainer — Workflow Automation Construction

DataloopTextData Collection

Developed automated data annotation workflows to remove inefficiency and inconsistency in parallel projects, enabling human-machine collaborative labeling loops. Pre-processed multi-modal inputs, mounted dynamic knowledge bases, and used CoT templates to generate initial annotations for review. • Pre-processed inputs using multi-modal techniques • Mounted dynamic knowledge bases for annotation support • Applied CoT templates to generate initial annotations • Configured backend interception for review to achieve 96%+ consistency

2024 - 2026

AI Trainer — Agent Optimization (SOP + CoT Annotation)

Function Calling

Annotated agent CoT and tool/code behavior to improve agent planning and reduce blind calls in complex inquiry-plus-measurement workflows. Built a thought-to-action-to-observation SOP and added self-checking scripts to label API and parameter defects for higher completion reliability. • Designed an SOP workflow from thought to action to observation • Annotated CoT from conditional retrieval to final consistency judgment • Labeled agent tool code and API/parameter defect cases • Increased complex decision completion rate by 40%

2024 - 2026

AI Trainer — CoT (Chain of Thought) Fine-Tuning

TextFine-tuning

Conducted chain-of-thought fine-tuning for quotation audit and contract compliance tasks where correct calculations still led to incorrect reasoning bases. Converted experts’ implicit industry experience into explicit trainable logic chains using a standardized thinking template and corrected-chain datasets. • Designed a logic breakdown standard (Background → Comparison → Re-examination → Judgment) • Built datasets with senior personnel supporting logic correction • Customized training templates to guide reasoning structure • Reduced misjudgment rate caused by reasoning errors by 60%

2024 - 2026

AI Trainer — RAG (Retrieval-Augmented Generation)

TextFine-tuning

Built RAG training datasets to address incorrect citations of retrieved materials in professional decoration scenarios. Used multi-candidate response alignment with multiple retrieval/reply models, then selected and manually rewrote citation fragments to ensure authority and traceability. • Introduced multi-candidate response alignment under the same prompt • Improved source coverage of key conclusions and guidance • Produced high-quality datasets via optimal selection and citation rewriting • Doubled coverage while ensuring authoritative, traceable process guidance

2024 - 2026

Show more

Education

Work History