Selected research prototypes, open-weight model collections, evaluation benchmarks, and full-stack AI applications developed to bridge theoretical NLP and reliable machine intelligence.
Open-weight ecosystem for computational humor. The Cognitive Synergy Framework runs six comedic personas in parallel, ranks funniest drafts, and distills them into efficient LoRA adapters across English, French, and Spanish. Explore at humorgen.pages.dev; benchmarks are in the HumorGen paper.
A live tournament-based leaderboard evaluating humor generation across state-of-the-art open and proprietary models with cognitive rubrics, hosted at humorrank-leaderboard.pages.dev. Benchmark write-up in the HumorRank paper.
End-to-end ML pipeline for multilabel detection of mental health conditions and cyberbullying using classical models and transformer-based embeddings. Published in PMLR 317:15–26, 2026 — Oral presentation & Best Paper Award at AAAI AIMedHealth 2026.
Specialized benchmark dataset addressing the "parametric knowledge gap" in AI economic analysis. Curated 4,309 multi-step QA pairs from 220 World Bank African economic reports (64,892 retrieval chunks) to benchmark temporal reasoning across leading models.
Designed, developed, and maintain the official research portal for the CMU-Africa NLP Research Group in Kigali. Features a dynamic paper showcase, member directory, collaborative project workspaces, and automated event tracking.
Enterprise logistics management system powered by an AI assistant that automates approvals, dispatches intelligent email drafts, resolves fleet exceptions, and syncs operational data bidirectionally with Google Sheets.
Official platform for the Carnegie Mellon University Africa Research Club — fostering interdisciplinary collaboration, conference submission tracking, academic resource curation, and member project spotlights.
Context-aware retrieval-augmented generation assistant designed for employee onboarding and policy inquiry across Slack, email, and web endpoints with strict document access boundaries and citation auditing.
Conversational urban navigation guide combining graph databases (Neo4j) and large language models to answer complex spatial, historical, and transit queries for the city of Kigali.
