KIXOProduct intelligence
Kixo
Ask your product a question. Turn the answer into a better decision.
Applied AI
Intelligence that earns its place in everyday products. From understanding an image to making complex information useful.
What applied means to us
Our AI work is grounded in products we operate. We take responsibility for the full lifecycle: data, training or prompting, evaluation, inference, monitoring and the decisions the model informs.
We build evaluation sets before selecting a model, verify model versions at deployment, and check input quality before expensive inference. Reliability, cost and latency are part of the product design.
Where it runs today
- Computer vision on skin imagery in SkinX: a calibrated ensemble with a quality gate in front of the pipeline and plain-language explanations generated by a language model.
- Document intelligence in Olbey: a vision-language pipeline turns bank statements from many banks and layouts into reconciled, structured transactions.
- Financial analytics in Olbey: transaction classification, behavioural insight detectors, anomaly flags, recurring-charge detection and cash-flow forecasting.
- Hybrid recommendation for a US social platform: large language models for understanding what people save, classical ranking for the feed, multi-provider orchestration with automatic fallback and full trace observability.
- A release-gated LLM coach in Intentions: an evaluation corpus with hard gates on safety and injection resistance runs before every release.
- An AI agent over product data in Kixo: the model selects from typed, audited actions and shows its reasoning trail.
- An AI coach in Football Pro: a retrieval-backed coach that answers players' questions around their training.
How we work with partners
Partner work starts with the problem and a way to measure success. We build the evaluation set, develop the model and create the product experience around it. The team stays involved through deployment and operation.