Shar

Key Skills

Software

Don't disclose

Other

Data Annotation Tech

Dataloop

Scale AI

Telus

Top Subject Matter

Bioacoustics and audio classification

Multimodal ML

object detection

Top Data Types

Audio

Image

Text

Top Task Types

Fine-tuning

Bounding Box

Segmentation

Classification

Entity (NER) Classification

Text Generation

Object Detection

RLHF

Computer Programming/Coding

Prompt + Response Writing (SFT)

Freelancer Overview

I'm just completing a masters in data science at the LSE ( London School of Economics and Political Science ). My specialisation has been in deep learning. Project 1: Fine-Tuning transformers using curriculum SpecAugment to mitigate class imbalance in bioacoustic classification. LSE Dissertation: Scalable scraping and dataset construction pipeline (object detection, embedding generation, zero-shot classification) Education includes Master of Science, London School of Economics and Political Science (LSE) (2025) and Bachelor of Science with Honors, King’s College London (2024). AI-training focus includes data types such as Audio and Image and labeling workflows including Fine-tuning, Evaluation, and Rating.

Labeling Experience

LSE Dissertation: Scalable scraping and dataset construction pipeline (object detection, embedding generation, zero-shot classification)

OtherImage

Built a scalable scraping and dataset construction pipeline combining object detection, embedding generation, and zero-shot classification for multimodal machine learning. Applied multimodal ML techniques using YOLOv8, CLIP, ResNet, and transformer-based models. Reported strong performance on vision tasks, indicating extensive model training and evaluation using constructed datasets. • Constructed datasets via scraping and automated annotation-like pipelines (detection, embeddings, zero-shot classification). • Trained and evaluated multimodal models including YOLOv8, CLIP, ResNet, and transformers. • Generated embeddings to support zero-shot classification workflows. • Achieved up to 91% macro-F1 across vision tasks through iterative evaluation.

2021 - 2024

Project 1: Fine-Tuning transformers using curriculum SpecAugment to mitigate class imbalance in bioacoustic classification

Don't discloseAudioFine-tuning

Developed a workflow for fine-tuning transformers using curriculum SpecAugment to address class imbalance in bioacoustic classification. The project involved preparing and curating training data for an audio ML task and validating model improvements through evaluation metrics. Focused on improving classification performance by enhancing robustness to imbalanced classes during training. • Bioacoustic classification dataset preparation and training pipeline work. • Curriculum SpecAugment strategy to mitigate class imbalance effects. • Transformer fine-tuning and performance evaluation for macro-F1 improvements. • Dataset/model iteration to achieve stronger results on vision/audio classification objectives.

2021 - 2024

Education

Work History