Fairness Research
Investigating bias in NLP models for sensitive content. My MSc dissertation reduced algorithmic disparity against Black women by 64%.
Explore projectsI'm a Chevening/Lemann Alumna (2024/2025) and Machine Learning Engineer at CloudWalk, working at the intersection of technology and social justice. I completed my MSc in AI with Distinction at the University of Essex, investigating intersectional bias in content moderation systems. Previously at IDB and Ernst & Young.
# Mitigating intersectional bias
from fairlearn.reductions import ExponentiatedGradient
from fairlearn.metrics import demographic_parity_difference
# Baseline accuracy gap: Black women -12.6%
baseline_gap = -0.126
# Apply in-processing mitigation
mitigator = ExponentiatedGradient(
estimator=base_model,
constraints=DemographicParity()
)
mitigator.fit(X, y, sensitive_features=groups)
# Result: Gap reduced to -4.6%
print("Bias reduction: 64%")
My work bridges the gap between technical implementation and ethical responsibility. From classifying harmful content to governing enterprise data, I build systems that protect people.
Investigating bias in NLP models for sensitive content. My MSc dissertation reduced algorithmic disparity against Black women by 64%.
Explore projectsProfessional experience in data loss prevention, GDPR/LGPD compliance, and AI security at Big Four firms and global fintechs.
View skillsChief Diversity Officer at Shaping Horizons. Building pathways for underrepresented groups in tech through Negritude no PhD.
See achievementsHow do classification algorithms inadvertently penalize LGBT and BIPOC creators? I'm developing fairness interventions using Fairlearn and custom NLP pipelines to fix this.
Applying computational methods to understand cultural patterns in digital spaces. My recent work analyzed 140,000+ films to map historical representation trends.
I'm always open to collaborating on projects that center equity, justice, and ethical AI. Whether you're a researcher, organizer, or just curious about algorithmic fairness — let's connect.