ML Engineer — TuneAfrica Recommendation Engine

About the role

TuneAfrica's recommendation engine must surface music that global platforms systematically fail to surface for African listeners — not just popular music, but the deep catalogue of emerging African artists whose music resonates culturally but lacks the algorithmic amplification that comes with being in Spotify's recommendation graph. You will build a recommendation system that is genuinely good at discovering African music for African listeners. The recommendation architecture combines three signal types: collaborative filtering on listening behaviour (users who listened to X also listened to Y), content-based features extracted from audio analysis (BPM, key, energy, genre characteristics from Essentia audio analysis library), and cultural context features (listener's region, preferred languages, social listening patterns). You will implement a matrix factorisation model for collaborative filtering, an audio embedding model using a fine-tuned music understanding transformer, and a contextual bandit for exploration-exploitation balancing in real-time recommendations. A key challenge is the cold-start problem for new and emerging artists. You will build a content-based fallback that can recommend new tracks with zero listening history based purely on audio features and metadata, and you will design the feedback collection UX (skip rate, repeat rate, playlist additions) to rapidly bootstrap collaborative signals for new content. You will build the real-time inference API serving recommendations with sub-100ms latency, the offline training pipeline running on a weekly schedule, the A/B testing framework for evaluating recommendation quality, and the diversity and cultural fairness metrics that ensure the system amplifies rather than homogenises African musical diversity.