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

1

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.

2

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

3

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