AI Researcher | Mastercard Foundation Scholar | Founder

Building inclusive AI for communities technology often forgets.

I'm Wanchi Lucia Yen, an AI researcher from Cameroon pursuing an MS in Engineering Artificial Intelligence at Carnegie Mellon University Africa. My work spans low-resource Bantu speech translation, Central African Sign Language recognition, and privacy-preserving AI systems for communities that deserve tools built with them.

80.39% CASL-TransSLR accuracy
#1 Zindi Skills2Job
56 students taught Python
Wanchi Lucia Yen portrait

Mission

record_voice_over

Low-Resource Speech

Designing speech translation and TTS systems for Bantu languages, with rigorous evaluation for high-stakes settings like education and healthcare.

accessibility_new

Accessible Communication

Building Central African Sign Language recognition systems that preserve privacy, reduce model footprint, and generalize to new signers.

80%
CASL-TransSLR

Pose-only Central African Sign Language recognition with signer-independent evaluation across 19 participants and a deployment-friendly 99MB footprint.

Read the research

Experience

Graduate Research Assistant

Vision & Language Lab, CMU Africa | 2025 - Present

Researching Project WAKANDA AI and CASL-TransSLR, building speech translation and sign language recognition systems for underserved African language communities.

  • Developed a privacy-preserving pose-only transformer for Central African Sign Language recognition, achieving 80.39% signer-independent accuracy.
  • Building PyTorch data pipelines, distributed training workflows, and reproducible experiments for low-resource Bantu speech translation.

Data Science Intern

Credable | Dubai, UAE / Remote

Built financial machine learning pipelines for income estimation and customer segmentation from mobile-money transaction behavior.

  • Engineered 77 behavioral features in PySpark and Databricks, reaching up to 0.967 R² for visible-income prediction and validating transaction-derived income proxies for affordability decisions.
  • Developed a cross-platform linkage pipeline for KashNow, Kwasa, and Airtel Money, connecting 5.87M customers to transaction behavior through identity validation and SHA-256 mobile matching.
  • Built a KMeans behavioral segmentation model with a 0.552 silhouette score and scored 7.35M+ customers into activity-based proxy tiers while supervised labels were being resolved.

AI Researcher

ECHO | 2025 - Present

Developing low-resource speech systems for ECHO, a multilingual conversational AI platform for African languages.

  • Built a 51-hour Kinyarwanda speech corpus from 28,263 unique phrases using large multimodal models to generate and curate synthetic training data at about $1.23/hour.
  • Fine-tuned and evaluated MMS and OuteTTS architectures for Kinyarwanda TTS, achieving 19.2% WER with OuteTTS-1B on human-recorded speech.
  • Researching model optimization for multilingual speech deployment on resource-constrained devices, reducing latency and compute requirements while preserving intelligibility.

Founder

YEN Pastries | 2019 - 2022

Founded and operated a pastry and catering venture, trained young women in practical skills, and grew an online community of more than 1,100 followers.

Education & Service

Carnegie Mellon University Africa

MS in Engineering Artificial Intelligence

GPA 3.9/4.0 | 2025 - 2027 | Mastercard Foundation Scholar

University of Buea

BTech in Computer Engineering

GPA 3.58/4.0 | 2022 - 2025

Community Teaching

Responsible AI & Python Workshops

Facilitated privacy-preserving AI training for 25 female students and helped teach Python to 56 students at Mahama Refugee Camp.

Research Tab

Research

Selected work across multimodal sign language recognition, low-resource speech, and responsible AI deployment.

Accepted | Deep Learning Indaba 2026

CASL-TransSLR: Central African Sign Language Recognition

First author: Wanchi Lucia Yen et al.

Developed a privacy-preserving transformer pipeline using pose-only representations, 225-D kinematic features, and signer-independent evaluation across 19 participants.

  • 80.39% signer-independent accuracy
  • 99MB model footprint for constrained devices
  • PyTorch, MediaPipe, W&B, Scikit-learn
  • View the Hugging Face model
WAKANDA AI | CMU Africa

Bantu Speech Translation Pipelines

Designing robust end-to-end speech translation systems that reduce cascading errors and improve reliability for African language technologies.

  • Distributed training and reproducibility workflows
  • Hyperparameter optimization and model evaluation
  • PyTorch, Hugging Face, NVIDIA A100, PSC
Competitions Won

Leaderboard Results

Machine learning competitions where rigorous validation, leaderboard discipline, and practical modeling came together.

Zindi certificate for AI4EAC Skills2Job first place
1st Place | 1/69

AI4EAC Skills2Job Intelligent Career Pathways Challenge

Built a ranking system to predict the five most relevant occupations from structured skill profiles, winning overall across competitors from eight African countries.

View competition
Zindi certificate for Yaounde mobility prediction third place
3rd Place | 3/53

2026 Yaoundé Mobility Prediction Hackathon

Predicted taxi ride times in Yaoundé using weather and trip data, placing third in the Zindi hackathon.

View competition
WiDS Global Datathon 2026 Top First Timer certificate
Top First Timer | June 2026

WiDS Worldwide Global Datathon 2026

Recognized as a Top First Timer for completing the 2026 Global Datathon Challenge in collaboration with Watch Duty.

View competition
emoji_events
1st Place | 2025

Tartan Innovate Hackathon, CMU Africa

Built ECHO, a multilingual conversational AI platform for African languages with voice and chat agents initially supporting English-Kinyarwanda and English-Swahili communication.

Let's create impact together.

Available for research collaborations, AI product work, keynote speaking, and educational workshops on inclusive AI and accessibility.

alternate_email

Direct Inquiries

wluciaye@andrew.cmu.edu
public

Based In

Pittsburgh, Pennsylvania | Worldwide