Liam Williams

Key Skills

Software

Clickworker

CloudFactory

Data Annotation Tech

LabelImg

Mindrift

Google Cloud Vertex AI

Labelbox

Top Subject Matter

No subject matter listed

Top Data Types

Document

Image

Text

Top Task Types

Bounding Box

Classification

Computer Programming Coding

Cuboid

Text Summarization

Freelancer Overview

Striving Data Labeling Professional with 4 years of experience converting raw images, documents, and text into high-quality training data for computer vision and NLP models. I have operated end-to-end annotation pipelines—bounding boxes, cuboids, classification, and text summarization—with Clickworker, CloudFactory, LabelImg, and Mindrift. By redesigning labelling pipelines and incorporating multi-stage QA controls, I decreased turnaround time by 70 %, lowered error rates to below 2 %, and helped clients accelerate model-training cycles by 30 %. In collaborative work with data-science teams, I translate visual requirements into intuitive annotation guidelines and deliver visual feedback that always yields clean datasets, attracting repeat business. My UCLA B.S. degree in Computer Science, Python (pandas, NumPy) proficiency, SQL, JavaScript, and Excel automation skills enable me to bridge technical expertise with rock-solid data hygiene—every label guaranteeing model performance.

Labeling Experience

Healthcare LLM Training Data Annotator

LabelboxImageBounding BoxObject Detection

Curated and annotated a 520 k-document corpus to build a clinical-grade LLM for a telehealth startup. Designed exhaustive NER schemas, built Prodigy recipes for rapid annotation, and used double-masked review + automated spaCy validation to ensure precision/recall at >0.97. Converted physician notes to de-identified Q&A pairs and brief patient-friendly summaries, allowing model deployment to be sped up by three months. The improvement process cut the turnaround in labelling by 60 % without lowering overall accuracy to 98 %+ and earning a follow-on contract worth $ 85k

Curated and annotated a 520 k-document corpus to build a clinical-grade LLM for a telehealth startup. Designed exhaustive NER schemas, built Prodigy recipes for rapid annotation, and used double-masked review + automated spaCy validation to ensure precision/recall at >0.97. Converted physician notes to de-identified Q&A pairs and brief patient-friendly summaries, allowing model deployment to be sped up by three months. The improvement process cut the turnaround in labelling by 60 % without lowering overall accuracy to 98 %+ and earning a follow-on contract worth $ 85k

2023

Autonomous Vehicle Perception Dataset Annotation Lead

Google Cloud Vertex AIImageBounding BoxPolygon

Managed a six-person annotation team to develop a 1.3-million-frame image-and-video dataset for an autonomous car company. Authored detailed labelling specifications, established CVAT workspaces, and designed Python QA checks that automatically identified outliers. The refreshed pipeline increased throughput by 70 %, the accuracy level remained at >98 %, and cut the client's model-training cycle by 30 %. Coverage included multi-class bounding boxes (traffic signs, pedestrians, vehicles) and pixel-exact polygons for lanes and drivable regions.

Managed a six-person annotation team to develop a 1.3-million-frame image-and-video dataset for an autonomous car company. Authored detailed labelling specifications, established CVAT workspaces, and designed Python QA checks that automatically identified outliers. The refreshed pipeline increased throughput by 70 %, the accuracy level remained at >98 %, and cut the client's model-training cycle by 30 %. Coverage included multi-class bounding boxes (traffic signs, pedestrians, vehicles) and pixel-exact polygons for lanes and drivable regions.

2022

Education

Work History