Mar – Jul 2025
Software Developer
Phoenixopia Solutions PLC, Addis Ababa
Built responsive client-side web applications in Next.js and Tailwind CSS, integrating REST APIs for dynamic data.
Sosna Worku Achamyeleh
MS student in Artificial Intelligence & Machine Learning
The George Washington University
I study whether deep learning models stay reliable when the data they meet in deployment no longer looks like the data they were trained on. My current work compares vision transformers and convolutional networks on chest X-ray classification under cross-institutional shift, looking at calibration, uncertainty, and whether explanation methods point to the right evidence.
I am a Black in Robotics + 3M Tuition Fellow, and I am applying to PhD programs in trustworthy computer vision, focused on the reliability and interpretability of vision and multimodal models.
Interests: computer vision, vision foundation models, multimodal AI, medical AI, representation learning, trustworthy AI.
Research
Reliability and Explainability of Vision Transformers Under Medical Distribution Shift: A Comparative Study on Chest X-Ray Classification
The ViT outperforms the CNN in distribution but degrades more sharply once the data shifts. Calibration breaks down for both, Monte Carlo dropout fails to flag out-of-distribution inputs, and attention rollout and GradCAM localize poorly under shift, raising the question of whether these explanations can be trusted when they matter most.
PDF coming soon Code coming soon
Experience
Research and Professional
Oct 2024 – Mar 2025
Junior Machine Learning Engineer (Volunteer)
Omdena San Jose Chapter, remote
Led 25 junior engineers on a predictive modeling project for coronary artery disease and heart failure risk using NHANES data, where XGBoost reached 88% accuracy and 0.85 F1.
Mar – Jun 2023
Software Engineer Intern
Escalate Africa Software Development S.C., Addis Ababa
Built a symptom-based hospital and doctor recommendation system with a 15-person team, developing ASP.NET Core REST APIs under Clean Architecture and test-driven development.
Teaching and Leadership
Nov 2023 – Mar 2025
Head of Communications Education
Africa to Silicon Valley (A2SV)
Designed and delivered a communication and technical interview training program for 205 university students from more than 20 African countries.
Mar – Aug 2025
Assistant Lecturer (Part-time), Computer Programming
Addis Ababa University
Taught lectures and labs on Python, object-oriented programming, and algorithms, and supervised undergraduate group projects.
Jul – Aug 2024
Teaching Assistant, Data Structures and Algorithms
AddisCoder
Supported 99 high school students across Ethiopia through mini-lectures and lab sessions.
Feb – Mar 2024
Back-end Development Instructor (Volunteer)
Ethioware EdTech Initiative
Trained 54 women across several African countries in backend development with C#, ASP.NET Core, and PostgreSQL.
Projects
Jul 2026
Face-Selective Visual Information and Deep Network Ablation
Neuromatch Academy
Built voxel-wise encoding models linking pretrained vision network features to fMRI responses from the Algonauts 2021 dataset, comparing AlexNet and a face-specialized ResNet-50 and testing whether targeted unit ablation removes face-selective information.
Jan – May 2026
Benchmarking Classical and Transformer NLP Models for Stress Detection
Compared Naive Bayes, TF-IDF logistic regression, and DistilBERT on the Dreaddit dataset across six stratified splits with McNemar, Wilcoxon, and Nadeau–Bengio tests. DistilBERT was strongest at 0.793 macro F1 and 0.880 ROC-AUC.
Sep 2023 – Jun 2024
Threat Detection in Spoken Amharic
Built and annotated a 5,800-clip Amharic speech dataset and compared CNN, RNN, and LSTM models on MFCC, chroma, and spectral features for real-time threat classification, with the CNN performing best.
Education
The George Washington University
M.S. in Artificial Intelligence and Machine Learning · Aug 2025 – May 2027 · GPA 3.78
Coursework in machine learning, neural networks and deep learning, trustworthy AI, algorithms, computer system architecture, and advanced software paradigms.
Addis Ababa University
B.S. in Software Engineering and Computing Technology · Oct 2019 – Jul 2024 · GPA 3.62, Great Distinction
Coursework in natural language processing, artificial intelligence, machine learning and big data, data structures and algorithms, and operating systems.
Additional Training
- Deep Learning Course, Neuromatch Academy, 2026.
- Data Structures and Algorithms Program, Africa to Silicon Valley, 2022 to 2023.
- Aspire Leaders Program, founded at Harvard University, 2023.
Honors and Presentations
- Black in Robotics + 3M Tuition Fellowship, GWU, awarded to one of the top two applicants.
- Best Poster Presentation (team award), Addis Ababa University Research Week 2024. Certificate
- Top 2% finalist, Aspire Leaders Program, 2023 cohort. Certificate
- Selected among the top 40 students in Ethiopia for Huawei Seeds for the Future, 2021. Certificate
- Poster, Benchmarking Classical and Transformer-Based NLP Models for Stress Detection, GWU SEAS Research and Development Showcase, April 2026.
- Poster, Threat Detection in Spoken Amharic Language, Addis Ababa University Research Week, May 2024.
Contact
I am glad to hear from prospective collaborators and PhD programs, and always happy to talk about medical imaging, explainability, and trustworthy AI.