RESEARCH
AI That Ships
We do research to solve problems—not to collect citations. Every project at Tusha Lab is aimed at production.
HOW WE THINK ABOUT RESEARCH
Tusha Lab exists at the boundary between frontier AI and emerging-market reality.
"What would it take to make this work—reliably, affordably, and at scale—in an environment with 3G connectivity, multilingual users, and cost constraints?"
We don't replicate Western AI research. We solve the problems that Western research ignores.
FOCUS AREAS
Multilingual Voice & Language AI
The Problem
Most speech recognition and NLU systems are trained on English and major European languages. African languages—and African-accented English—are underserved and misunderstood.
What we're building
ASR models optimized for local dialects
Understanding Nigerian, Kenyan, and South African accents accurately in noisy environments.
Low-resource language NLU
Developing understanding pipelines for Hausa, Yoruba, Swahili, and Pidgin without relying on massive pre-trained sets.
Multi-Agent Orchestration
The Problem
Enterprise operations involve coordinating multiple systems with long latency, API unreliability, and highly fragmented data schemas across legacy and modern tools.
What we're building
Resilient agent coordination in unreliable network states
Stateful handoffs across multi-modal enterprise systems
Computer Vision & Edge Inference
The Problem
Deploying computer vision locally in remote telecom towers requires offline-first inference on compute-constrained edge devices where cloud sync is episodic at best.
What we're building
Lightweight models for infrastructure anomaly detection
Efficient sync protocols for episodic connectivity