Welcome to the first-ever edition of Fleetyr’s In the Driver’s Seat. Our In the Driver’s Seat articles are a series of interviews with fleet professionals, industry leaders, and some of the new kids on the block from the fleet and transport tech world.
In the Driver’s Seat, today is Sanjar Bakhtiyar, the Founder and CEO of Solai, a US-based transport tech startup with Kazakhstani founders. Sanjar and the team at Solai are building a last-mile delivery optimization tool that can be used by fleet managers and fleet software providers to make the most out of their routing at a fraction of the cost of what the big players offer out there.
Sanjar’s on the cutting edge of AI technology and the quality already produced by Solai shows this. When not working Sanjar enjoys developing Deep Reinforcement Learning based AI bots – This bloke really is obsessed! Imagine what the team at Solai can create when Sanjar’s work is not working – it is pleasure.
What is something you are working on right now?
Here at Solai, we are trying to solve the fundamental decision-making problem for fleet managers and logistics dispatchers. Fleet managers must make all sorts of daily logistics decisions such as vehicle dispatching, scheduling, producing a vehicle routing plan, adjusting the routing plan due to order cancellations and late deliveries, and deciding what to do when the delivery truck crashes. These daily decisions heavily influence companies’ bottom line – specifically, operational expenses, customer service, and work productivity.
Fleet managers rely heavily on their experience to handle these edge cases. However, most decisions related to the last mile delivery are fundamentally mathematical optimization problems. Humans are good at finding patterns and making good/feasible decisions in near real-time. However, the decision can improve if the underlying mathematical problem is solved.
Since these optimization problems are very complex but managers require fast and reliable solutions in real-time, classical optimization solvers were infeasible and impractical for logistics. Luckily, with the advent of digitalization, IoT, and Artificial Intelligence, now we can automate those very complex logistics decisions and provide solutions in real-time. The recent academic research in Artificial Intelligence shows that AI Dispatchers can handle most, if not all, cases of logistics decision-making better than humans.
Our goal at Solai is to automate all possible logistics decisions so that the company’s last-mile delivery works like a swiss clock – reliably, predictably and, most importantly, efficiently.
What is something you cannot live without in your job?
I love talking to our potential customers, which are any logistics company and learning about their challenges and problems.
Logistics dispatching varies depending on the business. Every type of business has highly varying logistics objectives and operational / business constraints. Naturally, being an RnD-first AI startup, it is our passion to tackle different real-world logistics problems. It is our goal to automate all logistics decision-making. Naturally, such a goal comes with a lot of technical challenges, but we enjoy the challenge of solving problems that nobody before solved. Moreover, creating a digital twin of logistics and and warehouse operations is a lot of fun, and developing new Deep Reinforcement Learning dispatchers for those simulators is even more fun!
What is your dream vehicle?
My dream vehicle is one that has self-driving capability (Level 4-5 self-driving). In addition, they must be able to communicate with the nearby cars to make the best collective decisions. Cooperative decision-making and communication will allow for fewer traffic jams and, most importantly, avoid car crashes.
What is something you are proud of?
Before founding Solai, I was an AI research scientist at the South Korean university KAIST. There, for years I developed an AI dispatcher for various planning, scheduling and routing tasks, such as an AI scheduler for scheduling factory robots and an AI dispatcher for construction diggers, using Deep Reinforcement Learning. From an academic perspective, I developed a Deep Reinforcement Learning-based framework for solving scheduling and vehicle routing optimization problems, namely, Vehicle Routing Problem (VRP). Our AI solver works as good as industry-standard optimization and metaheuristic solvers on minor-sized VRP problems (under 100 orders and 10 vehicles). However, AI Solver shines when it comes to solving large-scale VRP problems (500-1000+ orders), outperforming optimization solvers such as Google OR-Tools by more than 15%. Moreover, AI Solver works 20x faster and is 1000x cheaper in terms of computational cost.
Our first client is an online grocery delivery company with an average of 1100 deliveries per day and a fleet of 46 vehicles. Our AI dispatcher helped to decrease total fuel consumption and mileage of delivery by 17.3% and enabled 100% on-time delivery while reducing the number of required trucks for delivery by 16.4%.
Fleet managers have a tough job – What are you guys doing to make their life easier?
Fleet managers’ jobs are indeed very tough. Starting from creating vehicle routes, work plans, and vehicle maintenance, they need to monitor the delivery operations and handle edge cases. There are many edge cases such as late deliveries, order cancellations, first-time delivery failures, and vehicle breakdowns. Last-mile delivery requires near real-time decision-making on ever-changing real-world operations.
Here at Solai, we develop an AI Dispatcher that will automate most, if not all, of the decisions required from the fleet manager. AI dispatcher is based on Deep Reinforcement Learning technology and can decide in real-time on vehicle route plans and real-time route adjustment depending on traffic jams, late deliveries and order cancellations. Additionally, the AI dispatcher continuously analyses new incoming orders and accordingly updates future routing plans.
What will the fleet industry look like ten years from now?
The trend in the industry shows that in 10 years, most logistics and operational processes will be automated. Starting from Amazon-style autonomous robots in the fulfilment centres, ending with autonomous trucks for freight forwarding and last-mile delivery. From the vehicle maintenance perspective, IoT for vehicle state monitoring and AI for analyzing anomalies will help preemptively repair vehicles before an actual breakdown, just like it is done now in advanced factories. On top of that, the whole autonomous fleet of trucks will be dispatched by a real-time AI Dispatcher who will analyze incoming orders and traffic jams and adjust the plan dynamically, according to continuously incoming new information about logistics operations.
If you think you have what it takes to be interviewed by the In the Driver’s Seat team at Fleetyr or you have suggestions on who we should be talking to, please contact us through our website.