Services / AI & Machine Learning
AI that pays for itself inside your operations
We build AI the boring way: pick one process where prediction or automation has a clear rand value, prove it in a small pilot, then scale what works. No moonshots, no demos that never reach production.
What ai & machine learning covers
Companies feel pressure to do something with AI, then fund pilots with no owner, no metric and no path to production. Eighteen months later there is a slide deck and no return. The fix is scoping AI like any other investment: one process, one metric, one decision point.
Machine learning models
Classification, scoring and anomaly detection trained on your data: credit risk, churn, fraud flags, quality control.
LLM assistants and copilots
Chat and copilot tools grounded in your documents and data, with guardrails, source citations and audit logs built in.
Document intelligence
Automated extraction and classification for invoices, contracts and claims, cutting manual capture from minutes to seconds.
Forecasting
Demand, cash flow and maintenance forecasts validated against your own history before anyone plans around them.
Recommendation and personalisation
Product and content recommendations that lift order value and retention, measured with proper holdout tests.
MLOps and deployment
Models shipped as monitored production services with retraining pipelines, not notebooks on someone's laptop.
What you walk away with
Every ai & machine learning engagement ends with assets you own outright: code, designs, documentation and the knowledge to run them. No black boxes, no lock-in.
- A feasibility assessment with an expected-return estimate
- A working pilot measured against an agreed baseline
- A production model deployed behind a documented API
- Monitoring for accuracy drift with retraining pipelines
- An honest evaluation report including where the model fails
- Handover training for the team who will live with it
Tools we reach for
Mature, well-documented technology your next hire will already know.
Modelling
Pythonsscikit-learn
PyTorchppandas
Hugging FaceLLM and retrieval
LangChainppgvector
Elasticsearch
RedisDeployment
Docker
FastAPI
PostgreSQL
GrafanaHow it runs
Find the money
A feasibility sprint scores your candidate processes by data readiness and rand impact, and picks one with a measurable baseline.
Results from this practice
Straight answers on ai & machine learning
What does an AI project cost?
The feasibility sprint is R80,000 to R120,000 and takes 3 weeks. Pilots run R150,000 to R500,000, and production deployment adds R200,000 to R600,000 depending on integration depth. Each stage has a go or no-go decision, so you never fund the next step on faith.
Do we have enough data for machine learning?
Often yes, and sometimes the answer is a simpler model than you expect. A few thousand labelled examples can support a useful classifier. The feasibility sprint answers this concretely for your data, and if the honest answer is no, we tell you what to start collecting.
How do you handle POPIA and data privacy with AI?
Personal data is minimised or anonymised before training, models run in your own cloud environment by default, and nothing is sent to third-party model providers without an explicit, written agreement on what leaves your boundary.
Will an LLM assistant make things up to our customers?
Unconstrained, yes. That is why we ground assistants in your verified documents, require source citations, restrict them to approved topics and log every exchange. For high-stakes answers we design human review into the flow rather than pretending the risk away.
Pairs well with
Put ai & machine learning to work in your business
A free 45-minute consultation with an engineer. You leave with options and honest numbers.