Role Overview
We build AI products for public health and early education deployed in low-resource field conditions across sub-Saharan Africa and South Asia. Models handle speech, audio, and vision on noisy, underrepresented data. You'll own the full ML pipeline from data preparation through training, evaluation, and deployment handoff. Small team means end-to-end ownership. Seeking solid fundamentals and desire to learn production ML beyond notebooks.
Responsibilities
- Prepare and clean field data (audio, images, annotations) with quality checks and dataset splits.
- Run training and fine-tuning experiments on speech, audio, and vision models; track and report results.
- Build evaluation sets and metrics reflecting real-world usage, not just held-out test splits.
- Perform error analysis: inspect audio/images to find failure patterns.
- Write clean, tested Python for data pipelines and deployable inference code.
- Build internal annotation and review tools with engineering when needed.
- Document experiments, datasets, and decisions for reproducibility.
- Follow data protection practices for health and child data; flag concerns.
Requirements
- 0-3 years experience or strong self-built ML project portfolio (internships, research, Kaggle, open source count).
- Solid Python with PyTorch (or comparable DL framework), NumPy, and pandas.
- Sound ML fundamentals: training/validation, overfitting, loss functions, learning curves.
- At least one end-to-end speech, audio, or computer vision project you can walk through.
- Basic Git, command line, and remote GPU job execution.
- Clear written communication for asynchronous, cross-time-zone coordination.
- Curiosity and rigour: verify results before trusting them; ask for help early.
Nice to Have
- Fine-tuning pretrained speech models (Whisper, wav2vec) or vision/OCR models.
- Model serving, quantization, or CPU/mobile inference experience.
- Experiment tracking tools and reproducible training setups.
- Low-resource language, child speech, or field-condition data familiarity.
- Interest in health/edtech or products for low-resource settings.