HLM Labs — where tomorrow gets tested
Our research arm runs experiments in frontier AI, local-language models, computer vision and edge computing — so what we deploy for clients is proven, not fashionable.
Innovation with receipts
The AI frontier moves monthly, and most of what's loudest doesn't survive contact with production. Labs exists to find out what actually works — on our infrastructure, our languages, our bandwidth and our budgets — before it carries a client's name.
Every engineer at HLM spends dedicated time in Labs each month. The best experiments graduate into client solutions and products; the failures teach us faster than any conference could.
Four bets we're making
Agentic systems
Multi-agent orchestration, guardrails and human-in-the-loop design for high-stakes workflows.
Low-resource NLP
Making language AI work for languages and dialects the big models underserve.
Practical computer vision
Vision systems that tolerate dust, glare and low-cost cameras — because real sites aren't labs.
Efficient & edge AI
Small models, quantisation and offline inference for low-power, low-connectivity deployment.
Experiments, pilots & open source
A public snapshot of the Labs bench — from early research to prototypes running in the field.
Vision QC
Low-cost camera stations for production-line defect detection, trained on-site in days and robust to variable factory lighting.
Shona & Ndebele NLP
Language models and speech interfaces for local languages, so digital services can serve everyone — not just English speakers.
Satellite Yield Signals
Fusing free satellite imagery with field records to forecast smallholder yields district by district, at near-zero marginal cost.
Agent Safety Harness
Our internal framework for permissioned, auditable AI agents — guardrails, approval flows and action logs — released for the community.
Edge AI on Solar
Running inference on low-power edge devices for sites with no grid and no connectivity — clinics, boreholes, remote depots.
PhishSim ZW
A phishing-simulation toolkit with localised lure templates, helping organisations train staff against the attacks they actually receive.
Open-source projects are published on our GitHub. Pilot programmes are open to selected partners — ask us about joining one.
Explore the ideas
Practical perspectives on data and intelligent automation.
Agentic AI: what autonomous agents mean for African business
How tool-using AI can support bounded business workflows, and the safeguards to consider before adoption.
ReadWhy your AI strategy is really a data strategy
Useful analytics and AI depend on relevant, well-governed data. Start with quality, access and a clear decision to improve.
ReadMachine learning or plain rules? Choosing the right tool honestly
Sometimes the honest answer is an IF statement. A practical framework for when ML earns its complexity — and when it doesn't.
ReadPartner with Labs
Universities, NGOs and businesses: co-run a pilot, sponsor research, or put a hard problem on our bench.