Laptop
AI platform for Finance

Finance

AI platform for Logistics & Transportation

Logistics & Transportation

AI platform for Autonomous Systems

Autonomous
Systems

AI platform for Decision-support simulations

Decision-support simulations

AI platform for Smart-cities

Smart-cities

Introducing VUKU, our AI
Platform for better decisions

Includes all tools & technologies for
decision-making in the real world

Platform components

Probabilistic
modelling

Reinforcement
learning

Multi-agent
systems

Cloud
platform

Spatio-temporal model

GP state space model

Deep GP

Tuneable AI

Hierarchical RL

Risk-sensitive RL

Model-based RL

Bounded rationality

Mechanism design

Demand matching

Resource allocation

Deployable decision-making system

Cloud learning system

VUKU is ideally suited for autonomous decision-making in complex, dynamic and uncertain environment such as financial markets, transportation & logistics, ridesharing, smart cities and robotics. It includes a comprehensive set of patent-pending AI tools and technologies to address varied requirements across all these industries.

The VUKU platform takes in environmental data to create and learn models and policies, and make decisions. It includes a growing portfolio of tools and technologies in the three main areas of focus at PROWLER.io - Probabilistic Modeling, Reinforcement Learning and Multi-Agent Systems.

The VUKU platform includes Probabilistic Modeling technologies such as Gaussian Process State Models for data-efficient modelling of complex, multi-dimensional environments. Spatio-temporal modelling techniques enable modelling of dynamic and uncertain environments like demand for rides in a city.

These probabilistic models are used for by Reinforcement Learning (RL) agents to plan strategies or policies for maximizing rewards in a given environment. Our Model-based RL enables orders of magnitude faster learning compared to alternative approaches. Using our patent-pending approach of adding the notion of bounded rationality in RL agents, applications such as autonomous systems are able to create policies and strategies that are more representative of the real world.

Most industrial applications are complex and require multiple decision-making agents, each either competing or collaborating with others. For example, each vehicle of a ride-sharing service in a city can be represented as an RL agent. The Multi-Agent System technologies for demand matching and resource allocation ensure optimal system-wide decisions across multiple agents.

read more

Our AI Platform for Decision-Making.
Flexible. Scalable. Reliable. Patented.

Animated Platform Diagram

Platform Overview

The VUKU™ platform (architecture diagram above) primarily consists of two main components - a decision-making component that is responsible for choosing the correct action to take in an environment, and a learning system that continuously learns new predictive models and policies in order to make the right decisions.

Using ride sharing company as an example of a use case, below is a description of how the various components of the platform can get used.

Prior to using the VUKU platform in production, the ride sharing company provides historical environment data that is passed in by the router to the environment data store (persistent storage), to be used by the probabilistic modeling based Model Learner. The Model Learner passes the model to the Reinforcement Learning (RL) agent(s) to be used for planning the most optimal strategy or policy to apply for the next observation.

Once the ride sharing company’s software component sends the environment data (ex. customer demand and location of various vehicles) and the reward from last ride via the Environment API, the Router routes this data to appropriate components of the platform. The Experience Store stores the observations in a persistent store, from where the Policy Learner learns and evolves the policies to be used by RL agents to make a decision. The most optimal decision across all RL agents is passed back to the ride sharing company’s software component waiting for a decision.

Flexible Deployment Model

VUKU has been designed and developed for industry use cases. It can provide decisions at scale, with high availability and fault tolerance. It offers flexible deployment architectures to suit the varied needs of different enterprise use cases. The decision-making and learning components can be deployed on different machines, even in different data centers. For customers, such as in finance, that want decisions in the shortest time possible, the decision-making system can be deployed in customer’s data center, the learning system can be fully managed in PROWLER.io’s data center. The platform can distribute workloads across many CPUs and GPUs to scale to meet the changes in demand.

Ready for Research, Development & Production

VUKU has been built ground-up for researchers and developers. As is typical in any AI system implementation, the platform supports experiment automation at scale, enables researchers and developers to monitor their experiments and perform post-hoc analysis. The platform tools also allow researchers and developers to record specific conditions necessary to accurately reproduce the results. This allows the developers to then build and configure the platform for production environment.