Prathyusha S.

Key Skills

Software

Remotasks

Scale AI

SuperAnnotate

Top Subject Matter

Voice-based security and wake-word robustness

AI fairness/governance for structured data

Demand forecasting data preparation and ML pipeline deployment

Top Data Types

Audio

Text

Image

Top Task Types

Function Calling

Data Collection

Classification

Object Detection

Question Answering

Text Summarization

RLHF

Fine-tuning

Computer Programming/Coding

Freelancer Overview

PhD Researcher – Personal Voice Assistant Security (wake word detection, jamming evaluation). Brings 12+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Keras, Docker, and FastAPI. Education includes Doctor of Philosophy, University College Cork (2025) and Master of Science, Vellore Institute of Technology (2015). AI-training focus includes data types such as Audio, Computer Code, and Programming and labeling workflows including Evaluation, Rating, and Function Calling.

Labeling Experience

Postdoctoral Researcher – Datamite (LLM fairness analysis framework)

Function Calling

Designed an agentic LLM fairness analysis system that integrates fairness tools and identifies sensitive attributes from structured datasets. Explained key bias metrics and fairness implications through tool-calling workflows within the LLM agent. Dockerized and deployed the framework across GPU clusters and cloud environments, and exposed it as a real-time FastAPI service for fairness analysis. • Integrated fairness tool calls for attribute identification and metric explanation • Operationalized fairness evaluation pipelines for structured data • Deployed across GPU clusters and cloud with Docker • Provided a FastAPI-based service for on-demand fairness analysis

2025 - Present

PhD Researcher – Personal Voice Assistant Security (wake word detection, jamming evaluation)

Audio

Developed and evaluated wake-word jamming attack frameworks using deep learning models to assess robustness under controlled acoustic conditions. Built simulated acoustic environments and multi-parameter simulation pipelines to test performance across varying power, timing, and noise conditions. Focused on adversarial training and transfer learning to improve model resilience for voice assistant security evaluation. • Trained and tuned Keras CNN/RNN models for wake word detection • Implemented adversarial training for noise resilience • Applied transfer learning for multi-wake word adaptation and classification • Designed simulation-based experiments to evaluate jamming effects

2021 - 2025

Supply Chain Demand Forecasting – Independent Research (data ETL + model training workflow)

Data Collection

Built an end-to-end ETL and forecasting ML workflow for supply chain demand prediction using large-scale datasets. Implemented ingestion, cleaning, transformation, and aggregation steps to prepare features for model training. Containerized and deployed the pipeline to AWS so predictions could be served via FastAPI for reproducible, production-grade MLOps. • Performed scalable PySpark ETL over supply chain data • Trained a scikit-learn model for demand prediction • Containerized the full pipeline for reproducible environments • Deployed to AWS and exposed predictions via FastAPI

2021

Education

Work History