ZYNTELLIGENT COMPANY LIMITED

BUSINESS INTELLIGENCE

Businesses that move fast — retail, agriculture and manufacturing alike — face a constant balancing act with inventory. Products typically have to sell within a season; beyond that, even steep discounts struggle to move them, and by law, unsold stock kept in storage can't simply be written off — it has to be sold, which keeps tying up capital and adding cost the longer it sits.

Minimizing leftover stock isn't as simple as setting a discount and waiting. Too steep a discount erodes both product value and the loyalty of full-price customers, who may turn to other brands; too shallow, and the goods still don't move in time. Every markdown decision carries trade-offs that ripple through the rest of the season.

With our background in AI research and game theory, we build business intelligence systems that let management foresee extensive what-if scenarios, envisioning the full scope of possible situations with real-time data across desktops, laptops and mobile devices — with thresholds tuned so decision-makers can explore consequences carefully, without being overwhelmed by information, before they act.

This is the same thinking behind our Personalized Data project for the Department of Agriculture Extension (DOAE), turning agricultural data into insight tailored to individual farmers rather than one-size-fits-all reporting.

DECISION SUPPORT SYSTEM

While typical application software deals with the routine tasks in an organization, more challenging and unforeseen problems — the kind whose outcome can decide a firm's success or failure — call for more advanced technology. Systems built for exactly this are known as decision support systems.

A hard disk production line is a good example. Each device must pass through the same machine three times, with 10 to 15 other processes in between whose timing varies unpredictably — and with no way to track what's already been processed, new devices get distributed into the line blindly. The result is a queue length that drifts unpredictably at every machine, and a very costly bottleneck.

Tackling a problem like this needs more than standard linear programming or queueing algorithms. With our background in AI research — a field whose original challenges lay in solving exactly this kind of complexity — we build systems that track what has been processed at each machine, predict how long the returning devices will take, and recommend how new devices should be distributed into the line.

APPLICATION SOFTWARE

Application software is actually our strength — we've done extensive work in this area for more than two decades. On the hardware side, our experience spans the days of mainframes and Unix workstations through PC-based and web-based applications to today's mobile platforms. On the software side, we've worked across technologies old and new: legacy COBOL, C, C++, Java, JSP, ASP and .NET.

We can build almost anything our customers need, in any size: ERP, CRM, supply chain, logistics and more — from scratch, or on top of existing open-source platforms such as Adempiere (a Java-based platform for ERP, supply chain and logistics) and vTiger (a PHP-based platform for CRM).

At the heart of how we build these systems is strict adherence to software engineering best practice. During the design phase, we make sure customers can see transparently what the system will do before development begins, and we put real effort into documentation — the most tedious part of the job, but one we know matters to the customers who rely on it.

Recent examples include a Task Management System (TMS) and User Management System (UMS) for Panasonic, a Tree Bank registry system for the Bank for Agriculture and Agricultural Cooperatives (BAAC), and an Enterprise Service Bus for integrating systems across an organization.

SIMULATION [sim.zyntelligent.com]

Simulation is used to study complex systems and to train personnel for situations where practicing in the real environment would be dangerous or costly — the same principle behind a flight simulator, which lets pilots build experience while keeping the risk of an unsafe flight to a minimum.

In the defense industry, we run simulations at both local and global scale. At local scale, individual entities — a soldier, a vehicle, a tank — are modelled as agents, each given a goal as simple as moving to a location. At global scale, agents coordinate and cooperate to jointly accomplish a shared mission: generating a plan, sensing the environment while carrying it out, and revising the plan as needed until the goal is achieved.

Beyond the simulation engine itself, the most convincing part of any simulation system is its visualization, which lets trainers and trainees observe the action as it unfolds and discuss it in detail during debriefing. At local scale we use VBS3 (Virtual Battlespace), a world-class defense simulation tool used by NATO forces and their allies, including Thailand and Australia; at global scale we use AIG for a wider range of simulation and visualization.

Delivered work includes 3D and crowd-ambience simulation on ArmA 3 and VBS3 for the Defence Technology Institute (DTI), a CBRN simulator built on Unreal Engine 5 for the Royal Thai Air Force (RTAF), and shooting simulators for the Department of Special Investigation (DSI, on VBS3) and the Royal Thai Army (on Unity).

HLA/RTI

We develop and integrate High Level Architecture (HLA)–based solutions specifically tailored for military simulation and defense-related training systems. Our expertise covers the design of distributed simulation architectures, RTI integration, Federation and Federate development, FOM modelling, time management, and reliable data exchange to ensure interoperability across complex system-of-systems environments.

In addition to custom engineering services, we provide a ready-to-use HLA Runtime Infrastructure solution, ZyntRTI, designed for operational deployment, testing and rapid integration. ZyntRTI enables simulation developers and system integrators to accelerate development cycles while maintaining compliance with HLA standards (IEEE Std 1516™-2010).

This same rigor sits behind our Tactical Data Link Simulation project, built on HLA/RTI for the Royal Thai Air Force — one example of the standards-compliant system-of-systems engineering we've delivered across Thailand's defense sector.

[Download ZyntRTI Data Sheet]

TAK · TEAM AWARENESS KIT

Team Awareness Kit (TAK) — also known as the Tactical Assault Kit — is a geospatial mapping and communication system built for situational awareness: tracking friendly forces in real time and supporting operational planning for military, government and public-safety users.

For the Royal Thai Air Force, we deployed Android Team Awareness Kit (ATAK) across cloud and on-site servers, connecting operations units over both internet and off-grid networks via the Cursor-on-Target (CoT) protocol — with live video feeds, voice, chat and drone telemetry brought together into a single common operational picture.

Our work spans the wider TAK ecosystem — ATAK for Android field units, WinTAK for command-post workstations, and TAK Server for backend federation, multi-user sync and data retention — extended through the plugin architecture to bring in mission-specific map overlays, sensor feeds and messaging on top of the open CoT data model.

VIDEO TRACKING

Unmanned aerial vehicles can be used for surveillance missions by means of video tracking — the process of estimating, over time, the location of one or more objects using information about objects of interest in the form of image pixels. By modelling the relationship between a target's appearance and its corresponding pixel values, a video tracker estimates the object's location; the main challenge is distinguishing the target from other objects in the scene as its appearance shifts.

Typical tracking works at three levels: low-level features such as colors, photometric invariants and gradients; mid-level features such as edges, interest points and regions; and high-level features such as background subtraction and object models. Each of these techniques has its own strength, performing well in particular settings and environments — but poorly outside them. This has been an active research area for us since 2012.

Our work advances these techniques by applying agent technology to draw out the best results from each one. Rather than tuning settings manually the way most researchers do, our agent-based system learns the best configuration for a given environment on its own, saving significant time, cost and manpower by working around the clock — then applies machine learning to find the best combination of techniques for the surveillance mission at hand.

OBJECT DETECTION

Building on our work in video tracking with traditional techniques, we've advanced further into object detection to help track objects in tough conditions — where the direction of both the UAV and its targets is unpredictable, the UAV's height and the targets' size are unstable, and image quality is often poor. In these environments, typical tracking techniques often fail to track accurately.

Object detection allows for more accurate tracking because it can recognize targeted objects regardless of changes in size or image quality. The system is trained on a set of example objects and maintains a concise knowledge base of their appearance.

Learning the best settings for each new environment remains the bottleneck — so here too, we deploy agent-based techniques to overcome it, then couple this object-detection-based tracking directly into the existing surveillance system.