Outcomes, not logos.

Engineers needed the right history, codes, documents, and photographs for rare, expensive faults. A multimodal retrieval workflow with reranking improved pilot recommendation evaluation and reduced the time spent retrieving information and drafting a remedy.

A major Japanese automaker needed better lead-to-salesperson matching and visibility into customer interactions. Personal LLM agents captured context during the work and recommended next talking points while salespeople retained the decision.

Every purchase request had added another vendor and another contract, and the same materials hid under different catalogue names. Five specialised agents normalised, researched, validated, optimised and sense-checked the options, and procurement kept the final decision.

The trader was losing repeat business to late follow-ups and losing hours to plates held up by mismatched test certificates. Two agents now capture and club leads by voice and check each plate against its certificate, with the owner approving what matters.

A global beauty business could forecast brick-and-mortar demand reasonably well, but D2C e-commerce demand was unstable. A planning model improved forecast accuracy, reduced bias, and helped teams reduce inventory imbalance, marketing waste, and late discounting.

A quality-control model was too slow for the line. Reframing vibration data as a signal-processing problem enabled edge inference without slowing production.

A multi-billion-dollar food exporter was losing margin because dealers could claim back poor-quality fruit. Batch-level shelf-life prediction helped the business decide which fruit to send first, where to send it, and how to reduce avoidable claims.

A telecom operation was over-provisioning field labor because faults were treated as random events. Decay modeling shifted maintenance from reactive dispatch to risk-based intervention.

A large professional-services organization was matching people to projects through manual HR coordination. A recommendation workflow helped HR and project managers find suitable resources faster, reduce allocation delays, and improve project fit.

A large US telecom provider could predict churn risk, but prediction alone did not create retention. The breakthrough came from connecting network-service issues to targeted intervention, using an agentic workflow to triage unhappy customers at scale.

A major automotive distributor was advised to classify which leads would buy. The better answer was to model the buying journey step by step, giving sales teams better prompts, finance context, and next-best-action guidance at each stage.

A US multi-brand fashion retailer had to commit stock 9–12 months ahead while balancing service and inventory across stores and online channels. Separate optimization methods addressed promotion price, order quantity, order timing, and DC allocation.

Demand, inventory and replenishment planning were built together for a 4,800-SKU operation. Service level improved; financial impact is still being measured.

Forecast accuracy improved from below 60% to 94% for a high-volume consumables client. This is a separate project from the beauty forecasting case.