Helios Logo



Jet up

We turn complex data into actionable strategy.

Utilizing an array of cutting edge machine learning, statistical modeling, and applied decision theory, our team delivers cutting edge artificial intelligence solutions for our clients.  Solutions are designed that adapt to their environment and learn from their experience.

Based on the research of Dr. Noah Silverman, our methods deliver real-world, measurable results that enable optimal insight and decision.



Uncertainty is inherent in any complex system. Traditional statistical and machine learning models generalize or average-out uncertainty,  ignoring many sources of variance and stochasticity in the data.  This can greatly compromise accuracy and hamper optimal decisions.

By using Probabilisitic Programming, Hierarchical Bayesian Modeling, and other advanced methods we are able to embrace, integrate, and understand uncertainty at every level of the process.  Probabilistic reasoning can then be utilized for decision making that fully quantifies and accounts for inherent system randomness.




We have extensive experience with development and testing of algorithmic trading models, custom derivative development, market micro-structure, order book dynamics, smart routing algorithms, portfolio optimization, and many other areas of fin-tech.


Dr. Silverman originally participated in a digital advertising “think thank” at SAMSI, and continued to research the area as part of his PhD work.   We have innovated multi-armed bandit algorithms for ad choice optimization, dynamic pricing models for PPC ROI maximization, and hierarchical models for attribution.


We have built models and algorithms for a variety of sports betting clients.  Our past work has been applied to horse racing, PGA golf, and cricket.  We continue to innovate new methods to quantify uncertainty in sports betting, and the application of portfolio theory, optimal bet sizing, and fund management.


Machine learning is rapidly improving many areas in healthcare and insurance. We’ve worked with both medical researchers and insurance companies, applying advanced statistical and machine learning solutions. Our core concept of quantifying uncertainty is especially applicable when dealing with the diversity of human disease and treatment.


Crypto-currencies, treated as an asset, are interesting markets and have significant profit potential for well constructed algorithmic trading. Separately, the technology underlying the blockchain enables trustless transactions and fully accountable transaction. We have extensive experience with both trading crypto as an asset and working with blockchain as a ledger system.


We have experience with natural language processing, document classification, subject abstraction, and document summarization.  Past projects utilized support vector machines and restricted boltzman machines  for term identification, classification, and as a decision engine.


Geo-Demographic Modeling

Development of statistical model of student enrollment for private, for-profit, educational institutions.  Incorporated multiple geographic and demographic factor to model and infer enrollment for proposed new campus locations.  Data from the US Census, bureau of labor statistics, and several other government data resources where collected and aggregated.  A hierarchical Bayesian zero-inflated poisson model was ultimately used.

Electricity Pricing

Realtime, dynamic modeling and estimation of day-ahead electricity demand and pricing.  Client is a major energy producer in the United States.  Dynamic Semi-parametric Factor Models were innovated along with custom time series trending and auto-regressive modeling were applied to real-time data flows from over 50 distinct data sources.

Dynamic Trading Algo

Private fund client had a unique technical trading indicator they had roughly conceptualized for trading specific commodity futures.  We created a dynamic, tradable indicator, then, developed a hybrid system of R and C++ to rapidly back-test trading strategies around  27 million variants of this new indicator.  The best choice delivered consistent and reliable returns during testing.


Development of predictive model to estimate true probabilities of horse race outcomes.  Variable extraction, derivation, and analysis.  Multidimensional combinatorial betting algorithm developed and implemented to exploit market inefficiencies in probabilities and dividends.

Optimal Portfolio

The Kelly Criterion is a well know and often used tool for bet or investment sizing.  Unfortunately, it quickly breaks down with multiple and or simultaneous outcomes for an event. Helios.ai developed a unique formulation and solution for Kelly Optimal portfolios covering complex event outcomes, multiple simultaneous outcomes, or multiple position choices.

Smart Order Routing

Designed and implemented fully automated system for filling client orders of FX and Bullion.  System was fully automated to receive and fill orders utilizing multiple liquidity providers in parallel.  Dynamic probability of fill events constantly re-calculated to drive order sending decisions.  Core technology implemented in Erlang for fully concurrent environmental awareness and actions.



Hierarchical Bayesian Models are used when data is available from several different levels or groups.  The hierarchical structure allows for analysis and understanding of complex multi-parameter structure and decisions.  MCMC methods are used to fit these complex models and return stochastic samples from the posterior distribution.  Utilizing Bayes theorem and specialized software, we are able to infer complex distributions from highly dimensional and hierarchical structured systems.


The probabilistic programming language approach within AI has the potential to fundamentally change the way we understand, design and build AI systems.  Many of the most innovative and useful probabilistic models published by the AI, machine learning, and statistics community far outstrip the representational capacity of standard techniques.  Probabilistic programming allows us to utilize the most innovative of models by treating data and coefficients as resulting from probabilistic distributions instead of point estimates.


Bayesian Decision Theory revolutionized how optimal decisions are made.  It is a statistical system that quantifies the tradeoff between various decisions, utilizing probabilities and costs.  An agent operating under Bayesian Decision Theory is constantly updating expectations, and subsequent decisions, as new data flows in.  This approach is based on quantifying the tradeoffs between various classification decisions using probability and the costs that accompany such decisions


A Gaussian process is a statistical model where observations occur in a continuous domain, such as time or space.  Every point in this continuous space is associated with a normally distributed random variable.  This method is useful for modeling complex, non-linear relationships while still accounting for noise in the data.  Predictions are not just point estimates, but also encompass uncertainty information.

Dr. Silverman on Linkedin: https://hk.linkedin.com/in/noahsilverman

To follow Dr. Silverman on Twitter: https://twitter.com/noah_phd


Contact Us

6 + 9 =