How a predictive AI architecture makes it possible to manage the complex and volatile logistics of prescription drug delivery and scale it to over 100,000 orders per day — as told by Rx2Go CTO Dmitry Chistyakov, the creator of the CPLOM algorithm.
Logistics in healthcare is one of the most critical and error-sensitive industries in existence. A single routing mistake can affect someone’s life and cost a patient their health. Every year in the United States, more than 1.5 million people are harmed by inefficient medication logistics, and the country’s entire healthcare system loses over $25.7 billion due to supply chain failures and drug distribution errors. To tackle this problem, Rx2Go — a company with over 15 million successful medication deliveries across more than a decade of operation — has achieved a breakthrough: its AI architecture has driven the delivery loss rate down to a record-low 0.01%. The technology’s creator, Dmitry Chistyakov, spoke with us about how experience across diverse industries helped him build a reliable AI-powered service for something as high-stakes as pharmaceutical delivery, how it expanded across 16 U.S. states, and how autonomous systems are replacing manual labor.
The Premier League of Logistics
“The American market is the premier league when it comes to complexity,” says Dmitry Chistyakov. “Here, medication delivery isn’t just dropping off a box — it’s an intricate web of insurance companies, regulators, and pharmacies. Operating on the East Coast, with its traffic density, means any delay triggers a cascading collapse. Either you automate everything down to the millimeter, or the market eats you alive.”
Over the course of a decade, Rx2Go — the company Dmitry works for — has completed more than 15 million successful deliveries. To navigate the challenges of operating within a heavily regulated system, the startup implemented an entirely new AI architecture: the CPLOM system, which brought the loss rate down to a record 0.01% — something virtually impossible to achieve through manual management. As CTO, Dmitry assembled a team of technical specialists, and one of his core priorities was accounting for the high level of accountability demanded by the U.S. pharmaceutical market.
“It’s a highly developed infrastructure — dense order volume, an extensive pharmacy network, and mature processes. At the same time, regulation isn’t centralized the way it is in many other countries: you have federal requirements, individual state requirements, internal hospital protocols, and pharmacy chain policies. All of these overlap. On top of that, you have vast distances and extremely expensive labor. You can’t just hire another 1,000 couriers when you fall behind — you’d hemorrhage money. Efficiency has to be mathematical.”
In conventional logistics, a mistake is an inconvenience to a customer. In medical logistics, it’s a risk to a patient’s health. Dmitry’s AI algorithm is designed to account for the regulatory requirements of each individual state and layer them on top of one another to produce an effective, functional delivery chain. Here’s how it was trained for that task:
“This was one of the most serious challenges,” says Dmitry. “We didn’t try to explain legislation to the AI in plain text — that’s precisely the path to errors. Instead, we went with an algorithmic hybrid approach. At its foundation is a deterministic meta-layer — a rigid framework of rules that cannot be violated under any circumstances: temperature requirements, driver licensing standards across different states, controlled substance handoff protocols.”
The key distinction is this: a conventional AI would suggest the fastest and cheapest route. When CPLOM operates, the meta-layer built by the technical team immediately overrides that suggestion the moment it conflicts with even a single clause of the law. Legal constraints became mathematical weights embedded inside the model, allowing the system to account for regulatory requirements faster and more accurately than an entire legal department. This proved especially impactful during and after the pandemic, when requirements were changing at an unusually rapid pace.
From Reactive Management to Minimal Delivery Loss
Through its AI architecture, Rx2Go achieved record-breaking results — the medication delivery loss rate dropped to under 0.01%. The biggest driver of this outcome was the shift from reactive to predictive management, Dmitry explains.
“We began treating delivery as a dynamic system of states, which allowed us to detect deviations 15 to 30 minutes before they actually occurred. The way it works is that we don’t rely on a single decision — we run it through multiple verification layers. Quality control is a good example. Previously, a team of dozens of people spent hours reviewing photos and geolocation data. After CPLOM was implemented, analysis time dropped to under 10 seconds. This didn’t just help us catch errors — it prevented them from cascading.”
What Medical Logistics Has in Common With Trading and Speech Analysis
Dmitry attributes CPLOM’s reliability to his 20 years of experience across wildly different areas of IT — the evolution of a systems architect. Back in the 2000s, he competed in the MetaQuotes World Trading Championship and ranked in the top three. That experience taught him how to build algorithms capable of making decisions in the midst of market chaos. Subsequent work in other industries added new layers of knowledge, but in practice, all of it was solving variations of the same fundamental problem.
“At Seopult, we automated the SEO industry and grew the startup from $100K to $100M in annual revenue. That’s where I learned to build systems that replace thousands of people with efficient algorithms. At WhenSpeak, my challenge was to build a platform for interactive audience engagement — tens of thousands of people simultaneously. When you have 25,000 users in a live session, the system cannot afford even a millisecond of latency. We ended up becoming market leaders, working with government agencies and pharmaceutical giants, and the startup was eventually acquired by a major bank for $8 million. Over time, it becomes clear that the core challenges are the same everywhere: scale, uncertainty, and complex interdependencies within the system. At Rx2Go, those approaches converged: the precision of financial trading combined with resilience under peak load.”
Trust Through Transparency and Global Expansion
The primary barrier to AI adoption in healthcare is what’s often called the “black box” problem — the inability to trace how the AI actually arrived at a decision. To overcome that barrier, Rx2Go bet on Explainable AI. Every decision made by the system became transparent to the client. Here, AI doesn’t replace oversight — it makes oversight mathematically formalized.
“The core fear is loss of control,” says Dmitry Chistyakov. “AI is perceived as something opaque, so our goal wasn’t simply to show results — it was to make the system explainable. We introduced dedicated personal assistant groups for each client, notification mechanisms for system actions, and real-time reporting. When a client understands why a decision was made and how reliable it is, their whole attitude changes.”
Rx2Go is now preparing to enter the Canadian and European markets. Despite differences in regulation, the CPLOM architecture remains universally applicable. Dmitry sees the primary challenge as preserving the resilience of the American model during rapid international scaling.
“In Europe, regulation is formally less stringent than in the U.S., and the market often doesn’t feel the need for change. We encountered the same thing in the U.S. during the COVID period, when regulations hadn’t yet taken shape and new approaches were met with caution. Over time, the solutions we were introducing as innovations became industry-wide standards — which worked in our favor as the market leader. Canada looks more promising; it tends to adopt American practices more readily.”
Comments are closed.