Capabilities / Artificial intelligence

Make the case for AI before the commitment.

A model demonstration does not establish a business case. The decision needs to account for integration, data access, review work, operating cost and who carries responsibility when the system fails.

I help executive and technical leaders decide where AI warrants investment and what to test before committing. This includes deciding which tasks need human review or deterministic controls.

Start with the task

What business outcome should change? What should the user be able to do? What happens when the model gets it wrong? How much can each completed task cost, including human review?

Those questions guide the choice of model, tools and architecture. Search, retrieval and agents are options to evaluate, not requirements to add by default.

Establish how the result can be verified

An explanation is not evidence on its own. Sources, explicit assumptions, reproducible calculations, tests and recorded actions make an AI system's work easier to inspect.

I help leaders assess the boundaries between the model and the rest of the system: what it can change, what must be deterministic and where a person needs to review the result.

At Semaverse.ai, I designed and built research agents, knowledge graphs and cross-source search, with tenant isolation and fine-grained permissions. The work included partner data integrations and a workflow connecting corporate PowerPoint templates, research and branded decks. These choices affect which sources an AI system can use, whose information it can reveal and how people can work with the result.

Use prototypes to inform strategy

Start with a strategic decision and identify the uncertainty that could change it. A focused prototype can test whether the proposed system will do the job and fit into the organisation's workflow. Agree what evidence is needed, how much work is justified and when to stop.

As a consultant to Camunda, I helped establish early AI product and technology work with prototypes that teams could evaluate and demonstrate. One assistant worked across documentation and software versions. Another used a smaller model, fine-tuned to help users write and correct expressions in Camunda’s language. I used a larger model to generate thousands of training and test examples.

The investment case must account for the reliability, permissions, failure handling and operating cost required in production. I help leaders assess those obligations alongside the prototype's results.

Separately, I built a custom tool-calling layer before the APIs I was using offered native function calling. That was independent experimentation, not part of the Camunda engagement.

Examine the effect on delivery

AI can accelerate coding and exploration while increasing the volume of work to review. A useful workflow makes ownership, verification and stopping conditions clear.

When a decision depends on an untested assumption, focused technical investigation can help resolve it. For example, does faster coding reduce overall delivery cost without weakening reliability?

Account for the economics

My 2023 RAG feasibility article examines the economics of an earlier Llamar product. It describes a working prototype whose operating costs made the proposed use case unviable.

Explore strategy and investment decisions or discuss your AI decision.