AI/NLP Engineer — AfriVote Sentiment Analysis

About the role

A consultation with ten thousand free-text submissions is useless to a government official who needs to understand public opinion in an afternoon. You will build the NLP pipeline that makes AfriVote's qualitative data actionable — transforming thousands of citizen voices into structured, meaningful insights that officials can act on. Your deliverables: a multilingual sentiment analysis model that classifies citizen submissions as supportive, neutral, or opposed to a proposed policy, and that works reliably in Swahili, English, and mixed Sheng; a thematic clustering system that automatically identifies the key themes and concerns raised across all submissions for a consultation; an automatic summary generator that produces a structured executive summary of public opinion with representative quotes for each theme; a topic modelling pipeline that discovers emergent issues not anticipated by the consultation designers; and a bias detection module that flags potential coordinated campaign submissions that may not represent organic public opinion. The NLP pipeline must be fast enough to provide near-real-time updates as submissions arrive — officials monitoring a live consultation should see sentiment and theme updates within minutes of submissions arriving. You will build the pipeline using Python, deploying models as FastAPI microservices with caching to handle throughput spikes. You will use Gemini and fine-tuned multilingual BERT variants for the core NLP tasks, with human-in-the-loop review workflows for edge cases. Experience with African language NLP, political text analysis, or multilingual sentiment analysis is highly desirable.