I spent three years building AI products.Now I am learning to choose which ones get built.

• Chemical manufacturing

  RAG diagnostic assistant

  Plant engineers lost hours tracing fault history across scattered documents.

  20% faster diagnostics, 400+ users, 70% adoption

• Industrial operations

  Predictive maintenance

  Equipment failures surfaced only after they had already stopped the line.

  15% less unplanned downtime

• IT support

  LLM ticket triage

  Support teams read and routed every incoming ticket by hand.

  40% less manual ticket handling

• Manufacturing analytics

  Line-level KPI dashboards

  Plant managers had no live view of where failures were concentrating.

  Real-time failure visibility across production lines

• Model strategy

  LLM selection framework

  Every use case reached for the largest model, whatever it cost to run.

  6+ models benchmarked on accuracy, latency and cost

• Enterprise search

  Document intelligence pipeline

  Answers were locked inside scanned documents no search index could reach.

  Document intelligence, embeddings and vector retrieval in production

• Client development

  Discovery to funded delivery

  Promising prototypes stalled before anyone would pay to productionise them.

  4 prototypes converted into funded engagements

Figures are the ones reported to the client at handover.