Amitha S.

Key Skills

Software

Clickworker

Data Annotation Tech

Toloka

Other

Top Subject Matter

No subject matter listed

Top Data Types

Document

Image

Text

Top Task Types

Evaluation/Rating

Fine-tuning

Prompt + Response Writing (SFT)

Text Generation

Text Summarization

Freelancer Overview

I have contributed to multiple Supervised Fine-Tuning (SFT) projects, including Vision SFT initiatives at Outlier, where I worked as a prompt engineer across mathematics, biology, and generalist domains. My contributions involved: Technical Contributions Data Curation and Labeling: I helped create high-quality labeled datasets essential for SFT workflows, ensuring that training examples represented optimal model behavior across multiple domains. This involved pairing appropriate help types with various image types in Vision SFT tasks. Prompt Engineering: I designed effective prompts that guided model outputs toward desired responses, helping to bridge the gap between pre-trained foundation models and specialized applications. Alignment Optimization: My work contributed to improving AI alignment with human preferences through careful crafting of training examples that demonstrated appropriate model responses in complex scenarios. Cross-Modal Supervision: For Vision SFT projects specifically, I tackled the challenge of creating localized image-text pairs that effectively taught models to generate appropriate responses to visual inputs. Quality Assurance: I evaluated model outputs against established benchmarks to ensure that fine-tuned models maintained high standards of accuracy and relevance within domain-specific contexts. Through these projects, I gained practical experience with the supervised learning paradigm that forms the basis of modern AI alignment techniques.

Labeling Experience

Vision SFT Prompt Engineering Project

OtherImageText GenerationText Summarization

Collaborated on a Vision SFT project at Outlier that involved creating and curating high-quality training examples pairing images with optimal AI responses. Engineered prompts for multi-modal supervision to teach models appropriate responses to visual inputs across diverse scenarios. Tasks included classifying different help types needed for various image categories, generating reference responses for model training, and ensuring alignment between visual content and textual outputs. Maintained stringent quality standards by following detailed guidelines to ensure training data accurately represented desired model behavior.

2024

Education

Work History