Akanksha B.

Key Skills

Freelancer Overview

I annotated and analyzed large-scale pedestrian behaviour datasets at Ahmedabad University using CORPIS software, applying textual and visual labelling techniques across 500+ hours of urban crossing footage to support model training and validation for autonomous vehicle HMI research. At Tech Mahindra's Maker's Lab, I expanded into multilingual data work, curating and pre-processing datasets for Bahasa and Malay, and structuring data pipelines specifically to optimize inputs for Supervised Fine-Tuning (SFT) models. I also implemented YOLOv8 for real-time traffic sign detection, building a custom scoring mechanism to improve localisation precision. I conducted official evaluation testing for Google's Gemma LLM, engineering adversarial and zero-shot prompts across 8 Indic languages and dialects to benchmark the Indus language model, which gave me direct experience in prompt engineering and model evaluation beyond labeling alone. Through my EqualView project, I trained ML models on ProPublica and Kaggle datasets for automated bias detection across race, gender, age, and LGBTQ dimensions, using NLP and FairLens to round out my data work with a fairness and bias-evaluation lens.

Labeling Experience

Education

Work History