Machine Learning consulting, Data Science consulting - if you've already decided that you need it, now comes the choice time. How do you find one that is worth your time and money?
1. Do I even need Machine Learning?
Truth be told, AI, Machine Learning or any other buzzword you may use, is not a holy grail and a cure for all the world's data problems. If you have been told otherwise, you've been lied to.
However, Machine Learning plugins to your company can increase the productivity of your employees by 2x or 3x as achieved my many of our clients. In this article, we will show some examples, how we did it.
As I like to say, an AI system is a friendly sidekick that solves a well-defined job, such as automatic article recommendation, detecting objects on an image and so on.
A good Machine Learning Consultants team will first assess if you need Machine Learning - if you're going to get an ROI.
I talk to a lot of people and quite often I talk them out of building an AI solution even though it might mean business to me. Especially early-stage startups don't need AI until they reach a critical mass of users.
Rule of thumb for "do I need AI" is: do you have a well-defined data processing problem that is important for you? Can you quantify the benefits (monetary or otherwise) of solving it?
You'll find a dose of inspiration here.
After you establish that building an AI solution is indeed for you, read on.
2. Is your specialty a match for my problem?
There are various domains where Machine Learning can be applied. Working with images and text requires different expertise and mindset. While some algorithms/techniques overlap and are domain agnostic (e.g. LSTM design can be used to work with both text and image sequence data), there is a substantial difference when you work with tabular data, text or images.
A good Machine Learning specialist for image processing should also understand the general challenges with computer vision - decide when to use OpenCV versus Deep Learning, how to observe underfitting/overfitting on the evaluation dataset and how to perform augmentation.
However, a solid NLP specialist needs to understand how grammars work, how text is composed, what are embeddings and how to build them, stemming. He also need to be familiar with recurrent deep learning algorithms, as well as boosting techniques.
3. Do you have general programming skills? Can you build, deploy and maintain a scalable solution on the Cloud?
Depends on your needs, but it's fairly likely after you build your Machine Learning model, you'll want to be able to plug it into your existing application. Deep Learning requires GPU to run efficiently, and scaling models requires an expertise
A lot of Machine Learning Consultants have a solid maths background, but not so much expertise in large-scale programming, or GPU programming. It's better to be a specialist than a generalist, but you still need someone to build end-to-end Machine Learning pipelines. You can read our Case Study to understand how we build such pipelines - along with code samples.
4. Do you know the state-of-the-art scientific literature? Do you know Deep Learning?
While Deep Learning isn't always applicable (arguably not even most of the case), it is a very important technique to consider especially when working with unstructured data.
We define unstructured data as easily understandable by the humans, but not by computers - like images or text. Deep Learning has delivered groundbreaking results and almost entirely solved tasks like image classification or object detection. It still requires serious expertise to work with.
5. Are you experienced commercially or academically?
Machine Learning is by far most heavily researched field of Computer Science at the moment. It has brought together world's sharpest minds - brightest mathematics and CS researchers for good.
While tough technical skills and analytical thinking are essential prerequisities, there's more than that. Not many ML scientists had an opportunity to work on analyzing, solving and successfully deploying a Machine Learning-based solution - to a client's success.
It's good to have a person or a team that approaches problems with combined scientific curiosity and business dexterity.
6. Do they take security & encryption seriously?
Processing sensitive data worth hundreds of thousands of dollars? Using government data in your project? What would it mean to you if that data was compromised? Better make sure that it's secure. Especially, if you are powered by Machine Learning hedge fund.
As always, the system is as strong as it's the weakest link. That means that you should do everything you can to protect it.
If one considers SSL as the only necessary requirement to achieve security, they should be dismissed immediately - it's just the baseline (along with VPN, database encryption, monitoring & auditing) - you need much more advanced methods.
Also, you will need strict, relentless people procedures to increase your chances of being safe.
When it comes to data,
there's simply no one-size-fits-all.
It's hard to find one person that can have all qualities, and that's why we assembled a team to solve world's most challenging data problems, and to a great success.
🕵🏻🕵🏻🕵🏻 Or, if you feel a bit Sherlock-y, dive here to learn more about Sigmoidal.