Example project, figures to be confirmed
On-premise AI cluster for a research institution
Research data that could not leave the country, on cloud GPU hours nobody could afford.
Client
A research institution, Nairobi
Sector
Schools and institutions
Services
AI agents & automation, GPU clusters & on-prem AI
Overnight
Job turnaround
Method
Median wall-clock time from submission to result, compared with the prior queue on shared cloud capacity.
Eight
GPUs on site
Method
Installed device count in the departmental cluster.
On premise
Data residency
Method
All research data now processed on hardware owned by the institution.
The story
Context
A research institution running computational work across several departments, under data-handling terms that ruled out overseas cloud providers.
The problem
Research data could not be sent to overseas cloud providers, and cloud GPU hours were unaffordable at the volume required. Jobs queued for days, and researchers had begun to shrink their experiments to fit the queue.
What we did
Workload sizing, an eight-GPU departmental cluster, a power and cooling upgrade, Slurm scheduling, private model endpoints and administrator training.
Sized the workload first
What they intended to run, converted into VRAM, throughput and node count, before any hardware was quoted.
Checked the building
Power draw and phase, UPS sizing, cooling load and floor loading. The cooling had to change before anything was racked.
Handed it over properly
Dashboards, alerting, a written runbook and administrator training, so the institution runs it without us.
The outcome
Jobs that previously queued for days now run overnight, on hardware the institution owns and staff administer themselves.
Next
A second node group, and shared model serving for the departments that are not yet on it.
Stack
More work
View all
Stock and dispatch system for a regional distributor
Four branches on separate spreadsheets, month-end taking a week.
One day
Month-end close
WhatsApp support agent for a service business
Two staff answering the same twelve questions all day, and nothing after hours.
Routine enquiries
Resolved without staff
Next step
Have a similar problem?
Tell us what it is costing you now, and we will tell you what it would take to fix.
We reply to every enquiry within one working day.