4 questions this podcast will answer:
01What are the steps involved in developing a deployable product?
02Top 5 reasons why Edge AI is being adopted in the products?
03What are the three significant roadblocks to moving from engineer-led proof-of-concept to large commercial deployment by business owners?
04How can the Proof-of-Concept cycle be broken?
This really is an exciting time, but delving into my team’s experience, I found that although cumulatively they have done a few thousand pocs, they only help bring about 150 embedded products into the market.
The question I ask is, why? What is this dealt by proof of concept? Why are so many proofs of concepts being created, and why do they provide so little value? And why do many products not taken to the commercial deployment stage?
For that, we need to first understand the process of product creation. Everything starts with the idea from an idea we move to a POC is nothing but an isolated exercise of validating the core features of the products and to check if it works.
From here, we move on to building the prototype; a prototype is where you create the product which has all the features, and also you test to make sure it performs as per end-user satisfaction. And from this point, we move to the pre-production device.
Here, Once the features have been confirmed in the prototype phase and tested, we focus on building a production-ready device, keeping in mind vendor’s, bill of materials, material availability, casing design aesthetics, And all those other parameters. Once this is done, we take the pre-prod device, and we get the necessary industry certifications before getting it to commercial deployment,
But then, this is how our ideal product development process works. But most of the time, in reality, development stops at the POC phase. Many enterprises actually do a lot of POC and not that many end products from a stakeholder point of view.
This is just a wastage of resources. Whilst Edge is the next big thing, we have seen a string of failed promises and under-delivery precisely because of this mismatch between the promised end product and the series of useless POCs, at least from the shareholders’ point of view, the engineers might beg to differ on that.
Before we step into this POC to production conundrum, let’s take a step back and ask ourselves why Edge is being adopted in the first place. Well, I would say it’s for the reasons of managing bandwidth and for having low latency; we need the prediction and insight engines to be deployed on the Edge.
Also, the given Edge AI solution should have high reliability and should be able to be deployed on scale whilst demonstrating a good ROI to the end customers.
So we understand for all these reasons and for providing increased privacy Edge AI, and Edge-based visual AI is a no-brainer, but then what’s stopping us from moving from POCs done by engineers to widespread commercial deployment by the business shareholders?
Well, actually, I split the reasons for this into three separate markets skill culture and business. The lack of the right skill team of engineers, designers, and product managers, along with insufficient data, is one of the few reasons of failures which can be placed in the skills market.
Also, an incompatible business culture and lack of tech visionaries in the top management is can be placed in the part of the culture bucket while lack of a demonstrable ROI and lack of business buying is in the business market.
Another reason I see why people move into Pocs is that it shows value easy but then moving it that value to commercial deployment is difficult. My engineering leads and I actually face the same problem while building vMeasure, which is actually an automated dimensioning system for warehouses whilst getting the initial accuracy to 80 was easy and pretty exciting because you are identifying a lot of new things right moving it to 98 repeatabilities with high accuracy of five emergent months of algorithm optimization AI model training and testing.
without high motivation, moving ai applications from POC to commercial deployment is very difficult as you have many variable factors which you need to handle to bring high accuracy and higher repeatability into your AI application;
Remember, if your solution prediction fails, most probably it is a human who gets hurt either financially or physically on the other end. Thus the complex nature of machine learning displays many people from moving away from the POC phase to the commercial deployment phase.
So the question is, how do we break this cycle? one of the ways to do it is to make sure that you have a clear understanding of all the problems you will face as you move to the prototyping phase in the POC stage itself.
Some of the major problems which I’ve seen happen to include availability of data and identification at which level of accuracy will the application be acceptable by humans for commercial deployment.
Based on my experience, I have understood that using off-the-shelf hardware and tunable algorithms helps us decrease the go-to-market time and get the product deployed faster in the marketplace.
Also, remember that a different skill set is needed to bring a product into the marketplace compared to researching to identify whether a particular feature set is viable or not.
It is always important to work with someone who can build your POC fast and identify the viability of the solution and quickly move the same into the POC stage.
At VisAI Labs, along with econ systems, our investor and partner, we have helped bring over 150 products into the marketplace. We have helped enterprises from designing the overall solution to providing embedded cameras and to providing the computer vision algorithms as well.
Now we also help them decrease the time to market for commercial deployment by building a set of algo accelerators and ready-to-deploy POC ports.
these POC boards are nothing but a combination of Nvidia jets and carrier boards with supported embedded cameras and deployable Edge-based uh computer machine algorithms, which can be converted into posts within four weeks or modified introduction ready devices within 12 weeks
Most firms nowadays work with Edge AI and computer vision product development experts who actually have this pre-made deployable set of hardware and software components to reduce risk and time to commercial deployment of their end products
So uh guys, uh if you have any doubts on how to incorporate Edge into your product or how to solve some issues in your Edge hardware or the software integration while building a visual AI solution, you can connect with us at firstname.lastname@example.org and we can help you take your product to commercial deployment stage faster and quicker
Thank you for tuning in to yet another episode of the podcast unraveling computer vision and hi for the real world. This is Alphonse signing off ciao!