What makes Xyonix different?
We are results driven. Many other firms prefer to work on an hourly basis. We, however, embrace risk and get paid for adding real value to your business, not just time spent. We offer fixed rate pricing throughout your engagement with us, so you know what business value you are going to get for your money.
We’re very deep in AI and ML. We’ve built, deployed and operated many state of the art systems over the years. We built the world’s first large scale music identification system, before AI was something you could say with a straight face which still serves billions of people 20 years later. We built the world’s first and largest deep grammatical NLP engine. We have cutting edge models in production analyzing thousands of surgeries globally to improve surgeons, listening to thousands of patient heartbeats, reading millions of documents, making high value predictions from business data, and the list goes on.
We understand business and technology — we haven’t always been consultants, we’ve built multiple successful venture backed companies. We’ve led large engineering organizations and we’ve brought strategic innovation to millions of users always using cutting edge technologies. We speak geek and biz fluently from your sharpest PhDs to your savviest board member, we can help shepherd your team through confusing times. We will offer you recommendations from many vantages including product, business strategy, development, data science, management and social. We understand not just how to build models, which is one small aspect of transforming your business with AI, but how to build the data gathering moat around your business, how to engineer your differentiating training data pipeline, and how to achieve human machine symbiosis so your humans can train and address model mistakes, and your AI systems can excel at what machines do best.
We know when to use AI — We’ve addressed hundreds of problems up close over the years. We have a knack for not just how to use AI, but when. Perhaps the most important challenge of businesses like yours is to identify when to use AI. Anyone can read some blogs and realize that predicting fraudulent credit card transactions is important, but many cases of using AI are much harder to detect. You might not realize what valuable insights can be gleaned from your millions of documents, images, videos, recordings or other business data. We have a proven methodology for asking questions, lots of questions, to uncover what hidden gems will truly transform your business.
We’re extremely agile, we like to get models into production in days or weeks not months or years. We know that excessive analysis can lead to paralysis. We are not academics, we like to build real working models and deploy them as fast as possible. We know we can cycle back and improve them, and we will and do, but we are most concerned that we prove out the key business hypothesis as fast as possible — often that means getting imperfect models into a complete business context so we can quickly assess whether improving them matters.
We’ve built massive systems at scale, so we know how to keep the pagers quiet at night. We’re so confident here, that we’ll actually carry the pagers so you don’t have to worry at all.
Who owns the work?
You own all work related to your project and you retain all rights to all of your data (including your data in any transformed state that it might wind up in).
See a short video discussing working with Xyonix and IP ownership:
See another short video about maintaining control of your project:
Who hosts your projects?
We host projects (via AWS) during the prototype stage (e.g. static sample data); this allows us to rapidly get a solution to you enabling your developers to start product integration. Our customers typically wish to quickly move to production, and we handle this seamlessly in our hosted environment. If you need to have your system hosted in your own facility for any reason, we are happy to accommodate as our platform is all built on easy to manage docker containers.
Do you ever mix customer data sets?
Who maintains your projects?
We take care of everything in the prototype stage and we typically maintain our production systems (we can do this very efficiently). If you want to transition the maintenance and development of the project to your own team we are happy to help you accomplish this.
How secure are your systems? What about PII and security?
Our founding technical partners have successfully developed projects for multiple government 3 letter agencies who have extreme security requirements. We can most likely work with you to provide you with a system with the technology and protocols to satisfy your needs. And in the case of severe PII restrictions we can work with completely anonymized data.
How can your projects be so fast and efficient?
Our founding technical partners have nearly two decades of ML and data science experience each, and, just as critically, much of this experience is a mix of building very large systems (experience with scale) and consulting experience helping small companies get projects started from scratch (experience with smaller budgets, shorter timelines, and bootstrapped projects).
We have leveraged this experience into producing our ML project platform, Mayetrix, that dramatically accelerates most machine learning projects.
Of course, faster projects means less risk and less cost for you.
We can often have initial systems running in weeks.
What about performance? How can I be sure your system will do what I am looking for?
No one can accurately predict exactly how well any given machine learning project will work. And, indeed, the performance of any live machine learning system is never static, it is always evolving.
This is why we always take a very iterative approach to our projects. Our first step is always to study your specific problem space and your goals: what business challenges are you trying to solve? What are business-relevant performance improvement metrics for you? And, critically, what does your data look like today?
If our initial study produces a promising result, our next step typically is to build quick proof-of-concept functional prototype. This will typically go a long way to answering the question “how well can this actually work?”
What business verticals do you serve?
This is a common question. However, we find that machine learning problems are better classified not by the vertical they serve but by the type of business problem they solve.
Many machine learning projects address one of the following three high level business problems:
1. Help me find more customers (and/or lower my customer acquisition costs)
2. Help me keep my customers for longer (anti-churn)
3. Help me sell my existing customers more stuff (upsell / recommendation engines)
We build systems to address all three of these goals.
Of course, we customize our projects to help suit your specific vertical, but more importantly, we customize our projects to suit the particular needs of your specific business and your specific data.
OK, I’m interested. What can I expect working with you?
We typically work using a very iterative, risk reducing process.
Our first step is always a simple initial conversation or two. Could we be a good mutual fit? Neither of us wants to work on projects that we don’t think can be successful. So we talk about your goals, your data, and your efforts so far. Mutually, we only want to engage when we can likely meaningfully improve on what you are already doing.
Our next step is typically a short paid engagement (usually a 1-3 week process). We look at a sample of your data and consider what you want to learn from the data and your desired business metrics. We develop a detailed plan for building a functional prototype and an initial production system. Once we have dug in a bit, do we still mutually believe we can provide business-relevant results for you, given the likely investment to proceed?
If we proceed, our third step is to build a prototype, or, more accurately, a series of functional prototypes to reduce risk and demonstrate initial model and system performance including APIs. We take as many reasonable shortcuts as we can (e.g. use static data) to as efficiently as possible demonstrate the answer the question: can the system perform?
If all goes well, we then seamlessly morph the prototype into a production system for you. Again, for efficiency and risk reduction, we always look to crawl, walk, run. Once we are live, we can continue to optimize and maintain the system for you or we can hand it off to your team at your option.
How much does this cost?
Our data science work is bespoke to your needs and always must provide your company with more value than it costs.
This is why we work closely with you at the start to make sure we understand what we need to accomplish for you in order to be successful to your business and to make sure we are a good mutual fit.
So the answer about costs is that (a) they always fit you and (b) they can vary widely.
Do you want to accelerate or expand the scope of a project once you see initial results that are promising? Then the pace and the costs can go up.
Do you want to immediately stop if an early study or prototype doesn’t provide promising results? Then the costs can stop right there.
We always work closely with you to make sure the costs, the risks and the value all match your goals.