Nairobi · Research Lab

We build AI systems that ship —
and teach teams to actually use them.

Wanx AI is a research lab that ships: enterprise software, model training & fine-tuning, data pipelines and agentic automation — plus hands-on AI staff training so the tools we build don't sit unused.

What we do

Four ways we work with a team

From production ML systems to getting a non-technical staff comfortable using AI day to day.

Enterprise software illustration

Enterprise Software

Production backends, data platforms and internal tools built to your workflow, not a template — designed to be maintained after we leave.

Model training and fine-tuning illustration

Model Training & Fine-tuning

Fine-tuning, evaluation harnesses and deployment for models that need to work on your data, not a benchmark.

Data pipelines and agentic automation illustration

Data Pipelines & Agentic Automation

Multi-agent systems and validation pipelines that process data reliably at scale — the unglamorous infrastructure that makes AI usable.

AI staff training illustration

AI Staff Training

Practical, role-specific training that gets a team using AI tools in their actual daily work — not a generic slide deck.

0%
Cost cut on a client data pipeline
0
Active partner engagements
0M
Records processed in production
0
Service lines, one lab
Selected work

A few things we've built

Multi-agent reinforcement learning research and production agentic tooling — a look at how we approach hard problems.

AIDAW agentic media engine terminal output
Agentic Systems

AIDAW

An agentic, petabyte-scale media engine — a supervisor agent coordinates dedicated ffmpeg, search and data agents, each backed by its own LLM, to process media at scale from a single CLI.

Multi-agent reinforcement learning football simulation
Multi-Agent RL

MARL Football

Teams of agents learn coordinated play from scratch via self-play — passing, positioning and marking emerge without being hand-coded.

Multi-agent reinforcement learning Mahjong simulation
Multi-Agent RL

MARL Mahjong

Agents trained under imperfect information and a large combinatorial action space — a harder coordination and inference problem than most game environments.

Who we work with

Partners across sectors

From auto-service SACCOs to NGOs to AI companies — same two things every time: working software, and a team that can use it.

Rightsify
Action Foundation
Sparkscar Auto Tune Limited
Mickey Justar Auto Fit Center
Mema Auto Assessors Ltd
Planet Q Productions
Rightsify
Action Foundation
Sparkscar Auto Tune Limited
Mickey Justar Auto Fit Center
Mema Auto Assessors Ltd
Planet Q Productions
Rightsify
Action Foundation
Sparkscar Auto Tune Limited
Mickey Justar Auto Fit Center
Mema Auto Assessors Ltd
Planet Q Productions
Rightsify
Action Foundation
Sparkscar Auto Tune Limited
Mickey Justar Auto Fit Center
Mema Auto Assessors Ltd
Planet Q Productions
Rightsify
Action Foundation
Sparkscar Auto Tune Limited
Mickey Justar Auto Fit Center
Mema Auto Assessors Ltd
Planet Q Productions
Rightsify
Action Foundation
Sparkscar Auto Tune Limited
Mickey Justar Auto Fit Center
Mema Auto Assessors Ltd
Planet Q Productions

See all partners & case studies →

Have a system to build, or a team to train?

Tell us what you're working with — we'll tell you honestly whether AI is the right tool for it.

Get in touch