Skip to content
vibermacher

We build AI systems that capture how your best people think and work.

Bespoke tools, trained models, and the judgment to know which is worth building.

What.

Find where judgment lives.

Your most valuable work isn't written down. It lives in how your best people decide — which client to take on, which fund to back, which signal to trust. Start there.

  • A portfolio manager's selection instinct.
  • A founder's client read.
  • A senior advisor's risk call.

Systematise it.

Build tools and train models that reproduce quality of output of key individuals. Judgment becomes infrastructure.

  • A tool that writes investment rationale as your lead manager.
  • A model trained on years of top salesperson outreach.
  • A looped system that researches, drafts, edits and polishes.

Built around you.

Shaped to your domain, your voice, and how you actually work. Not a generic wrapper — built for your use.

  • Your data, your terminology, your workflow.
  • Output indistinguishable from the original — not a polished approximation.

How.

  1. Understand.

    Spend time with the people whose judgment matters, to understand where the data that captures their thinking lives. Most of that isn't written down anywhere.

  2. Curate.

    Find examples that represent their best work, not their average output. This is where most AI projects fail — the model is only as good as the data set.

  3. Systematise.

    Build the tool or train the model. Then make sure it holds up against the original. Debug, improve, iterate.

Curation is the work. Deciding what counts as a good example of someone's considered judgment — versus their rushed output — can't be automated. That's where we live.

Work.

Wealth management

Blended-fund selection tool

Ingests live portfolio holdings and generates investment rationale in the voice of the firm's Portfolio Manager. The output isn't a template — it reflects portfolio construction logic and explanation specific to the individual.

Township labour markets

WhatsApp-native gig-matching platform

Designed around how informal workers build trust and work. Full stack: Claude-powered intent classifier, matching engine, React dashboard, POPIA-compliant data architecture with HMAC-tokenised PII.

Senior executive archive

Voice model

Fine-tuned on years of interviews, speeches, and recorded thinking using transcription, speaker diarisation, and LoRA / QLoRA on local models. Produces tweets, short-form content, and long-form drafts from the subject's considered output, not surface patterns.

Personal knowledge infrastructure

Multi-agent orchestration system

A routing orchestrator over specialised sub-agents handling research, web-clipping, synthesis, and vault filing — running as a private second-brain system. Telegram interface, live on local hardware, processing and filing continuously.

Approach.

We pair years in finance and strategy with hands-on AI engineering — the combination that lets us tell which judgment is worth systematising, and then build it efficiently and effectively.

Engineers struggle with business context; strategists can't ship. We do both.

Contact.

Tell us what your best people do that you can't scale.

A short note is enough to start.