Build
AI is the current multiplier on strategy, workflow, measurement, training, and review. Build that layer before it is packaged, and it can be an advantage. Wait for the vendor version, and you are renting the status quo.
§01
Most engagements start with a workflow under pressure — slow, expensive, opaque, inconsistent, or held together by a few people. Four kinds of work recur. They mix; serious problems usually need more than one.
Strategy
Start from the business problem, not the tool. Map where AI can multiply speed, judgment, quality, training, or margin; separate build, buy, pilot, defer, and forbid; leave with the capabilities worth holding inside the organization.
Build
Turn a costly, opaque process into an inspectable AI-supported workflow. Source-grounded documentation, intake summaries, review queues, internal knowledge retrieval, and quality checks. Reduce cycle time. Leave a review trail people can use.
Infrastructure
Build the measurement layer the program depends on: events, outputs, review signals, dashboards, data structures. The operational evidence that governs quality, risk, revenue, and decisions about what to change next.
Evaluation
Vendor-claim reviews, AI readiness audits, and research-engineering support for studies or systems that won't fit off-the-shelf tools. The point is to know what the system actually does before procurement, rollout, or publication depends on it.
Specific starting points
Audit
A fast review of workflows, data access, staff capability, risk constraints, and vendor exposure. Output: where to pilot, what to stop, what not to automate, and what capability should be built inside.
Workshop
A cohort, leadership-team workshop, or organization-specific program for operators who need shared judgment before a larger build makes sense.
View programme →Pilot
A source-grounded workflow for reports, intake summaries, clinical documentation, internal memos, funder responses, or executive briefings, with review points and a usable trail.
Review
Independent evaluation of vendor AI claims against clinical, behavior-analytic, operational, data, and governance standards before purchase or rollout.
§02
Phase A
Workflow inventory, current evidence, decision points, integrity signals already in use.
Phase B
What changes, what stays human, what evidence is required, what error modes are intolerable.
Phase C
A working system with an inspectable trail. Real data, real review, real failure cases.
Phase D
Runbook, integrity checks, the internal capability to keep operating it.
§03
§04
I have run the kind of organizations the work on this page is meant for, and built the systems behind them.
Co-founder of Knospe-Lerncenter, which became Germany's largest ABA therapy provider — eight facilities, eighty-plus staff, ABA alongside speech, physical, and occupational therapy. Later CEO of Ausblick Therapie, where locations and headcount tripled inside a year.
Researcher by training, with peer-reviewed work on decision-making in the experimental analysis of behavior and neuroscience. Since 2024, presenting and teaching on AI in applied work. Daily user of frontier AI and multi-agent systems, including the open-source ClawSuite Relay.
The work on this page is what those years of operating and research turn into when an organization needs AI to actually change how it runs. Background and CV →
§05
If AI needs to improve real work — not add another layer of tools — write to me.
Direct
I'll tell you honestly whether this is the kind of thing I can help with — usually within a few business days.
