Serve your own LLM API
Deploy a gated endpoint on H200 pools in Nairobi.
A launch pad for African ideas.Build in the lab. Ship when it's right.
From agentic workflows to production endpoints, prototype with agents, train on local GPUs, gate quality, and ship endpoints.
Iterate in the lab
Build full-scale AI systems.
Built with Ai-lab
Fine-tune on regional corpus
LoRA on A100s with eval gates before promote.
Run a support agent locally
Alomana + your fine-tuned model, all on Ai-lab.
Batch transcribe at scale
Whisper jobs across warm GPU pools — pay per second.
Multi-node pretrain
Gang-schedule 8× H200 nodes with one CLI command.
Background research agent
Long-running tasks with lab storage and audit logs.
Full-stack AI capability
From advisory to factory floors — one lab surface for research, engineering, infrastructure, and agentic AI on African compute.
06 offerings · 01 platform
01
AI Research
Long-running training, ablations, and benchmark suites on dedicated H100 and H200 pools.
02
AI Advisory
Architecture reviews, model selection, and compliance planning for regulated African operators.
03
AI Factory as a Service
End-to-end factory capacity — data ingest, train, eval, and promote — without building your own MLOps stack.
04
AI Infrastructure
Bare-metal and virtual GPU tenancy with NVLink-aware scheduling and project isolation in Nairobi.
05
AI Engineering
Fine-tuning, distillation, and serving pipelines with gated quality and latency checks before production.
06
Agentic AI
Multi-agent automation on sovereign models — document intelligence, fraud, and ops agents via Alomana.
Scalable AI-infra at the core of
capacity.sh
ailab capacity request \
--sku h200 \
--nodes 4 \
--quota rnd-llmNVIDIA V100
Baselines and batch inference
Memory
16 / 32 GB HBM2
Peak
112 TFLOPS
NVIDIA A100
Fine-tuning and multi-node jobs
Memory
40 / 80 GB HBM2e
Peak
312 TFLOPS
NVIDIA H100
Frontier training and serving
Memory
80 GB HBM3
Peak
989 TFLOPS
NVIDIA H200
Long-context and high-memory workloads
Memory
141 GB HBM3e
Peak
989 TFLOPS
8 × H200 ≈ 672,000 tokens/s peak decode
Four stages to production
ailab@gpu-node-01: ~/lab
queued
live
running
—
complete
—
Live cluster visibility
Cluster heat
ops.sh
ailab status rnd-finetune-17
# GPUs 64×H200 · 1.84M tok/s
# stage train · ETA 02:14:08rnd-finetune-17 · live
Stage timeline
Ingestdone
Trainlive
Evalqueued
Servequeued
GPUs
64 × H200
Tokens/s
1.84M
ETA
02:14:08
Industries we work in
Financial Services
Fraud, loans, and KYC — data stays in Kenya.
Healthcare
Genomics, ops AI, and hospital workflows.
Manufacturing
Vision on the line. Predictive maintenance.
Security
CCTV and IoT fusion into live intel.
Common questions
Lab nodes in under six minutes. Studio pools are warm.