During the Cloud First era, the big technology challenges became obvious over time, such as building distributed systems to survive failures and designing distributed databases to ensure data integrity during a network partition. This is what brought amazing engineers to big search engines, à la Google (a terrific investor in UiPath via CapitalG), and to big born-on-the-web companies, i.e., Netflix. These specific tech challenges even became the prominent battle for talent on the HBO hit series, Silicon Valley.
While one traditional tech company recently referred to Robotic Process Automation (RPA) as screen scraping in a post it is far from that. With advancements in machine learning (ML), computer vision, application interfaces, and process orchestration for scheduling at our core already available, these are among the key reasons why UiPath is witnessing simultaneously very fast growth and high customer enthusiasm.
I recently sat down with PD Singh, VP of Artificial Intelligence (AI) at UiPath, to get a feel for the big challenges we are addressing to deliver highly resilient and secure robotic operations and AI at massive scale.
We’re quickly moving toward a time when mission-critical robotic operations will impact every employee, customer and supplier, and all will be working together to deliver extraordinary competitive advantages.
Watch PD’s presentation from UiPath India DevConf 2019 here for a deep dive.
Here are some of the biggest challenges ahead in RPA operations that we’re tackling at UiPath:
Distributed machine learning execution engine: Robots execute workflow processes and machine learning over arbitrary numbers of machines in an enterprise or across a value-chain. So, basic distributed enterprise platform functionality is table stakes. And some of the interesting problems here include state persistence of these robots (and the apps they run on top of) over tens or hundreds of thousands of virtual instances.
Probabilistic threshold discovery: RPA is inherently a logistic system with deterministic steps in it. Introduction of probabilistic steps (AI models) within RPA poses a big challenge for the specification and tracking of the correct execution semantics. Think, if a probabilistic system is 70% confident that the prediction is true positive, should it let the rest of the process (e.g. of creating a support ticket for a repair technician) execute? For non-life threatening and non-mission critical processes, we might be able to learn the confidence threshold levels by using human-in-the-loop and reinforcement learning.
Process understanding: Our customer success managers (CSMs) are often asked to help customers find the next biggest opportunity for automation. And as good as they are, we believe our machines may actually be able to answer this question on their own by analyzing user patterns. This can happen by recording and understanding human processes and then analyzing to improve long-running processes across multiple information systems and humans.
Visual understanding: The better a robot can see and digest elements of a user interface (UI), the more processes that can be automated. We’re investing in human-like recognition of UI elements by balancing a detection and mining capability that can work in a closed-loop system even when data is too sparse and so huge that it can overwhelm a thin client system.
Document understanding: New sources of unstructured data will need insight and high accuracy to be included in an effective automation. These forms of data range from signature and handwriting verification to analyzing multiple screencast videos to understanding and classifying many images simultaneously in term of relevance, ranking, and relationships. Check out another great video from UiPath India DevConf 2019 with UiPath’s Vitan Moliya on Document Processing Today with Intelligent OCR.
Conversational understanding: Natural language processing (NLP) can help automation robots understand the sentiment of text, chat, and voice inputs. We need a way to dynamically constrain the dictionaries of the natural language understanding (NLU) system to be able to understand the human inputs better in very specific verticals of processes. An RPA conversational programming capability will facilitate the rapid programming, training, and instruction of robot operations and decisions in English.
These are just a few of the skills that robots will acquire in the long run. As new use cases emerge, new skills will emerge. This is a virtuous cycle.
We are already seeing partners develop business models for industry or domain specialized algorithms that can be deployed via RPA. The same goes for customers who already have the data models, but have struggled to deploy them. RPA is making it easier for customers to drag and drop AI right into their workflows.
Our development centers in Bangalore, Bellevue, and Bucharest are in full swing, hiring some of the most curious and talented engineers seeking to use AI to do amazing things. In the 'automation first' era, we will solve (along with our partners) the next generation of incredible tech challenges for distributed, mission-critical robotic operations and machine learning at massive scale, much like Google and Netflix innovated to scale the cloud first era.
BuiltinSeattle just recognized UiPath among the 5 Seattle best tech companies where work matters most. While we concentrate development in these cities, we are a passionate and productive, highly distributed company operating in 31 cities across 19 countries. Madrona Venture Group, based in Seattle, is one of our key investors helping us achieve this amazing accomplishment. . . thank you, S. Somasegar!
Join UiPath and you will work alongside Param Kahlon, Marius Tirca, Andra Ciorici, Lavinia-Andreea Cojocaru, Bogdan Ripa, PD Singh, Prakash Thekkatte, Palak Kadakia, Mihai Moise, Nic Surpantanu, Boris Krumrey, Munil Shah and hundreds of engineers, all equally passionate to tackle the scale challenges to deliver on the real, long-term potential of AI in our customers’ digital business operations. JOIN US!!!