Skip to content
NewBranch

Six services that work as one system.

Private AI is mostly data work. A knowledge assistant needs clean records; an agent needs clear permissions. We plan them together so nothing is built twice.

01 / 06

Data foundations

AI is only as good as the records behind it. We connect your systems, remove duplicates, agree definitions with each department and build one store your models and your people can both rely on.

Typical deliverables

  • Inventory of systems, files and who owns each one
  • Pipelines from your systems into one governed store
  • Shared definitions for customers, products and key figures
  • Data quality checks with alerts to a named owner

Outcomes

  • One answer per question, not four
  • Records ready for AI without another clean-up
  • Fewer spreadsheets passed around by email
02 / 06

Private model hosting

We choose the right model for each job, open-weight or commercial under a private agreement, and deploy it in your own environment and region. Prompts, documents and answers stay inside your boundary.

Typical deliverables

  • Model shortlist tested on your own data
  • Deployment in your cloud account or on your hardware
  • Access control, logging and cost budgets
  • Fallback and upgrade plan for every model

Outcomes

  • No company data sent to public chat tools
  • Running costs you can predict per month
  • Freedom to change model without a rebuild
03 / 06

Company knowledge

We turn policies, contracts, manuals and records into a searchable company memory. Every answer cites the document it came from, and every person only sees what they are allowed to see.

Typical deliverables

  • Document ingestion from file shares, email and line-of-business systems
  • Search and retrieval tuned on real questions from your staff
  • Permissions that follow your existing access rules
  • An evaluation set that measures answer accuracy

Outcomes

  • Answers in seconds that used to take an afternoon
  • New staff productive in days, not weeks
  • A source behind every answer
04 / 06

Agents and automation

We automate the repeatable steps around your data: intake, extraction, follow-ups, reconciliation and reporting. Agents act inside clear limits and hand anything unusual to a person.

Typical deliverables

  • Workflow study with the people who do the work today
  • Agents connected to your email, CRM and finance systems
  • Approval steps and limits for every action
  • A log of every action an agent took and why

Outcomes

  • Hours of manual work removed each week
  • Fewer errors from re-keying the same data
  • A full record of what was done automatically
05 / 06

Security and governance

We set the policies, controls and records that let your risk committee, auditors and regulators sign off. Built around POPIA, GDPR where it applies, and your own information-security standards.

Typical deliverables

  • AI use policy and acceptable-use guidance for staff
  • Data classification and access review
  • Risk register and model evaluation reports
  • Incident plan for when an answer is wrong

Outcomes

  • A written answer for every audit question about AI
  • No sensitive data reaching tools you did not approve
  • Staff who know what they may and may not use
06 / 06

Interfaces and adoption

A private model is only useful if people use it. We design the screens, the place in the workflow and the training, then watch real use and remove the steps that slow people down.

Typical deliverables

  • Interviews and task study with the people who will use it
  • Assistant inside the tools your team already works in
  • Prototypes tested with real users before building
  • Training sessions and short written guides

Outcomes

  • Use that grows after launch instead of fading
  • Fewer questions to the same three experts
  • Feedback that goes straight into improvements

Three ways to work together.

Pick by how long the work runs and how much of your own team is involved. Many clients start with an assessment and move to a project.

ModelWhat it isTypical lengthBilling
Data and AI assessmentWe review your systems, data and risks, test two or three use cases on your own records and hand you a written plan with costs.Typical length: 2 to 3 weeksBilling: Fixed fee
ProjectA defined scope with a start, an end and a fixed set of deliverables. Good for a data platform, a knowledge system or a first agent.Typical length: 6 to 16 weeksBilling: Fixed price per phase
RetainerA set number of days each month to run, evaluate and improve what we built together, with a named engineer.Typical length: 12 months, renewableBilling: Monthly

Not sure where to start?

Most companies start with one question they cannot answer quickly today. Describe it and we will tell you which two services it touches and what to do first.