The basics
 

The Rainbird Studio provides a visual interface that enables domain experts to model expertise in formal representations of logic called “knowledge maps”. These knowledge maps are superior to decision trees because they are holistic (knowledge maps can answer questions beyond the scope of the original design), scalable, quicker to build and maintain and can handle uncertainty. 

These knowledge maps can be (optionally) connected to data to enable human-like reasoning and determination to be made at a large scale. When there is insufficient data, Rainbird can bring a human-into-the-loop and gather data from humans at run-time. 

In technical terms, maps are graph models which combine (probabilistic) logic with links to (potentially uncertain) data so that Rainbird’s inference engine can provide answers to queries. The no code visual mapping interface enables experts to express weights certainties and optionality over rules and rule conditions as well as data. The outcome is high-quality, high-speed nuanced outcomes which are evidence-based, accompanied with levels of certainty and an explainable audit trail that describes how each decision was reacted. The outcome typically outperform the expert that builds the model in terms of quality as well as speed.

The maps modelled in Rainbird are typically a formal representation of human-derived logic. Rainbird is well established in a range of sectors Financial Services (fraud, financial crime, credit decisioning), Insurance (FNoL, claims) and Professional Services / Accounting (tax, audit, on-boarding) as well as Telecoms (support, diagnostic). 

Rainbird is most valuable where it is important to be able to explain how each decision was reached. This is what differentiates Rainbird’s human-derived algorithms from machine-learnt algorithms which are statistical in nature and therefore not readily interpretable. 

There are a number of ways that Rainbird can evolve through use to strengthen outcomes. 

Discovery

Rainbird is able to discover and infer new facts about the world through being used. The Author of each model has discretion of how Rainbird should remember and re-use learnt facts. 

For example, you may teach Rainbird that People speak Languages, Countries have national Languages, and People are born in Countries. You may provide a rule that says that people are likely to speak the national language of the country they are born in and assign a certainty to that rule. If you query the model and ask, “What Language does a particular Person speak, Rainbird will dynamically construct questions to discover the data it needs to  return an answer. The Author decides if and how these facts should be remembered and re-used, and in what context. In this examples this may include discovering new people, countries, languages and the relationships between them. 

Changing weights and certainties

Rainbird knowledge maps can be engineered to take feedback on an answer it gives, such that future inferences will take that feedback into account.  This allows you to build models that change their output in the face of specific feedback they’ve learnt - but there is a trade off between this and human-built knowledge maps. Rainbird is powerful at applying “approved logic" at scale. Models that take feedback to influence outcomes may compromise the value to be had in building a model that is understood and approved. 

Rainbird and Machine Learning (ML)

It is often better to combine Rainbird’s inference engine with a machine-learnt algorithm that may be trained to make a predictions over data. This can deliver ML-derived predictions matched with human-derived judgements to deliver high quality auditable decisions. 

The ML aspect can be built in any number of ways, but the Rainbird platform natively* has functionality to create machine-learnt algorithms and combine them with human derived models to create hybrid outcomes.

This hybrid knowledge mapping/ML approach can give superior results with partial explainability and partial learning. 

*available in beta as of May 2021