AI That Plans Deliveries in 5 Seconds: 4 Real-World Logistics Innovations

CHAPTER 1
Beyond Calculation: How Far Has AI Come in Logistics Decision-Making?

"Can AI independently decide where, whom, and how to deliver pharmaceutical products?"

When asked this, Professor Jin-Gyu Park, CEO of OmeletAI and faculty at KAIST’s Department of Industrial & Systems Engineering, responds without hesitation:

"Absolutely."

This isn’t an empty promise. Park not only asserts the claim, he demonstrates it. An AI system generating thousands of pharmaceutical delivery plans in just five seconds—then dynamically updating those plans in response to changing road traffic, absent delivery drivers, or new last-minute orders. In short: machines perform the calculations people can’t, while leaving human judgment intact.

In the logistics field, this structure isn’t entirely foreign. For decades, logistics has relied on a mix of instinct and experience. Field operators know that "Route A always clogs up," or "Drivers avoid that pharmacy because of its long waiting time." These aren’t just anecdotes—they’re patterns of tacit knowledge that haven’t been easily digitized.

But what if AI could learn that intuition? More precisely, what if it could absorb the decision-making logic behind human expertise? Park describes this shift as a transition from operations-based logistics to decision-based logistics. It’s a world where AI becomes more than a calculator—it becomes a partner in decisions.

To put it plainly: every day, logistics professionals ask themselves, "How do I ship today’s orders in the fastest, most efficient, least resource-intensive way possible?" OmeletAI offers a compelling answer.

In this article, we explore four innovations where AI is already reshaping the future of logistics: pharmaceutical distribution, food delivery, parcel box optimization, and even unmanned maritime vessels. We also examine Park’s view on what ultimately determines success—or failure—when AI meets the real world.

CHAPTER 2
Pharmaceutical Logistics: AI Planning Delivery to 10,000 Pharmacies in 5 Seconds

If you manage pharmaceutical logistics for thousands of pharmacies, you may be skeptical about AI handling delivery route planning. But Park has turned that vision into a working system—backed by thousands of real-life data points. And more than the numbers, it’s the feeling that convinces users. A process that once took over a day is now done in 5 seconds.

More importantly, the difference lies in how the AI handles uncertainty.

Imagine delivering medical products to 10,000 pharmacies across Seoul and Gyeonggi Province. Simple distance calculations won’t cut it. You must consider how many stops a driver can handle, who gets which route, vehicle capacity, product constraints, and strict delivery time windows. Even after hours of human effort, the result won’t be perfect.

But OmeletAI completes the task in seconds—and more critically, it responds to disruptions. If orders change or traffic builds, the AI recalculates instantly.

According to Park, the system isn’t just an algorithm; it’s a learning-based decision model that incorporates years of logistics expertise. The tacit knowledge—"that route always clogs," or "that pharmacy is easy to work with"—is embedded in the model.

"The ability to respond instantly to unexpected events like last-minute order changes, traffic jams, or driver absence—that’s what defines performance. And that directly impacts customer satisfaction and operational efficiency." — Jin-Gyu Park, CEO of OmeletAI

The results speak volumes. In a pharmaceutical delivery test involving 1,300 pharmacies and 10 drivers:

Total travel distance decreased by 20%

Total task time reduced by 45%

Beyond cost savings, this translated into a more reliable healthcare supply chain—a critical public good.

Park is clear: AI is not replacing logistics professionals. Instead, "AI handles repetitive, mechanical decisions; humans focus on strategy and exceptions." This is the human-AI collaboration model Park envisions.

In a field long dominated by manual decision-making, this shift is significant. Because logistics is not just about moving boxes—it’s about making decisions. And AI, it turns out, may be better at that than we imagined.

CHAPTER 3
Food Delivery: Teaching AI the Logic of Batch Dispatch

When OmeletAI first introduced its route suggestions to food delivery riders, many were skeptical. "Is this really the best route?" they asked. The AI's paths often looked strange—unfamiliar alleys, unexpected turns—deviating from the riders’ hard-earned instincts.

But once they tried it, their opinions changed. Particularly during batch deliveries—where riders pick up multiple orders—the AI’s recommendations saved considerable time.

Batch delivery is a logistics nightmare. Orders arrive unpredictably, traffic patterns fluctuate constantly, and maintaining food temperature and freshness adds extra constraints.

Yet OmeletAI’s system learns and adapts. It evaluates each batch scenario in real-time: rider availability, weather, traffic, container type, food category, and more. And it delivers its decisions in under three seconds.

Imagine 1,000 simultaneous food orders in downtown Seoul. Each one has different prep times, destinations, packaging, and delivery windows. Coordinating that with 30 riders? Humanly impossible.

That’s where OmeletAI’s learning-based decision system kicks in. It evaluates all these factors, makes decisions instantly, and—crucially—learns from its mistakes. Park even compares the process to the AI "going through adolescence," shaping its character through feedback.

“There’s no perfect system. But we’ve found that having AI generate the base plan, and letting human riders adjust as needed, works best. Their adjustments feed back into the system, helping the AI learn and improve.” — Jin-Gyu Park

Real-world results:

Delivery distances reduced by 13%

Labor costs decreased by 6.8%

Customer wait time reduced by 22%

This isn’t just about efficiency. It’s about reducing rider stress, improving customer satisfaction, and boosting platform profitability. A rare triple win.

AI isn’t flawless. But it doesn’t have to be. Park’s model is about symbiosis: AI proposes, humans adapt, AI improves.

CHAPTER 4
Parcel Packaging Optimization: Cracking the 3D Puzzle with AI

With new packaging regulations kicking in from 2024, excessive packaging has become both an environmental and operational headache. Too little padding risks product damage. Too much, and you violate policy—and drive up costs.

Parcel boxes are money. More boxes mean more packaging cost and higher transportation spend. What if AI could solve that?

OmeletAI collaborated with a major parcel carrier to optimize parcel box usage. This wasn’t about fitting items in boxes—it was solving a complex 3D puzzle.

Each product has a unique size and shape. Only a limited range of box types is available. The goal? Maximize filling rate while ensuring product safety. OmeletAI’s solution combined 3D bin packing algorithms with machine learning to:

Recommend the most suitable box in just 0.8 milliseconds

Reduce simulation time from 2 weeks to 25 minutes

Identify the 6 most efficient box types out of 15 variants

All with just a barcode scan.

Operational staff don’t need to understand the math. They scan an item, and the system instantly recommends the best-fitting box. Training time plummets; productivity soars.

In the pilot case, the results were dramatic:

Packaging material costs reduced by ₩1.2 billion/year

Transport cost savings of ₩700 million/year

Filling rate improved by over 5%

Not just a cost-saving tool—this became a compliance-friendly ESG asset.

CHAPTER 5
Autonomous Vessels: Swarm Intelligence and the Future of Distributed Logistics

“Why limit AI to trucks and warehouses?” Park asks. That’s how he introduces his work on Unmanned Surface Vessels (USVs)—developed in partnership with Hanwha Systems for naval mine clearance.

On the surface, this doesn’t look like logistics. But Park insists: “The system is essentially a floating warehouse of robots.”

This is Multi-Agent Collaboration—a distributed AI system where each vessel independently assesses its surroundings, avoids collisions, and completes assigned tasks without central command. Park calls it Decentralized Decision-Making.

Now apply this to logistics:

Dozens of Autonomous Mobile Robots (AMRs) navigating a warehouse

Thousands of last-mile delivery vehicles operating in real-time

The lesson? Swarm Intelligence. A system where agents act independently, yet in harmony.

This isn’t about individual optimization—it’s about collective decision evolution.

Park’s takeaway is clear: technology validated in defense can—and should—transfer to logistics. Because logistics, at its core, is just a battlefield of decisions.

CHAPTER 6
Why AI in Logistics Succeeds—or Fails

Despite all the hype, not every AI implementation succeeds. Some companies see huge returns. Others get stuck in pilot purgatory. Why?

Park offers a decisive answer: “It’s not the technology—it’s how you define and execute the problem.”

5 Pillars of Success:

Clear Problem Definition

Don’t say, “Let’s use AI.” Say, “Let’s reduce route distance by 15%.”

Phased Rollout

Avoid enterprise-wide disruption. Start small, test, expand.

Domain Expert Integration

Logistics operators must help train the model. Without them, AI lacks context.

High-Quality Data Management

Garbage in, garbage out. Collect, clean, and validate.

Operational Flexibility

AI suggests. People adapt. Hybrid decision-making works best—especially early on.

4 Common Failure Traps:

Overinflated Expectations

Insufficient Data

Lack of Workflow Integration

Neglecting Change Management

“AI is not magic—it’s a tool. Its success in logistics depends not on the algorithm, but on how deeply it understands the field problem—and how well it collaborates with the people who live it.” — Jin-Gyu Park

Who is Jin-Gyu Park?

Jin-Gyu Park is CEO of OmeletAI and Associate Professor of Industrial & Systems Engineering at KAIST. He holds a B.S. in Architectural Engineering from Seoul National University and an M.S. and Ph.D. in Civil and Electrical Engineering from Stanford University. His work focuses on decision-centric AI systems that blend human expertise with machine optimization. His team’s innovations span pharmaceutical delivery, food logistics, parcel packaging, and unmanned maritime systems.

His signature concept—the Learning Decision System—reimagines AI not just as a tool for automation, but as a partner in strategic, adaptive decision-making in complex logistics environments.