Driving Progress with AI: Softacom’s New Venture in Computer Vision and Video Processing

Softacom has been deeply engaged in AI solutions for years, working with frameworks like OpenAI, and the ones from Google, and Microsoft. This includes generative AI, predictive neural networks, and computer vision.

But why AI? Why focus on computer vision and video processing?

In this article, Serge Pilko, Founder and CEO of Softacom, answers the most common questions that we receive on the AI transformation services and shares the company’s journey behind the new direction, the creation of the Softacom AI Lab, and how our solutions address real-world business challenges across industries.

Why a New Direction in AI?

During 16 years of focusing on Delphi, .Net and legacy software migration, Softacom has been privileged to work with various AI solutions and frameworks to support our clients and their projects.

This includes generative AI leveraging tools from OpenAI, Google, and Microsoft (both cloud-based and local models like Ollama), as well as the development and training of predictive neural networks.

For instance, a few years ago at the conference, I presented approaches for developing and training a deep learning predictive neural network to forecast wildfires in Spain. Additionally, we’ve worked on solutions involving computer vision.

Through Softacom’s projects, we’ve encountered a wide range of tools, approaches, and third-party products, always striving to integrate and enhance them. Over time, our focus naturally gravitated toward tasks involving the analysis and processing of photo and video content. Instead of creating custom expert systems, we began utilizing trained neural network models. This shift led us to establish computer vision as a dedicated direction and to invest in internal development, enabling us to offer unique, cutting-edge solutions to our clients.

What is Softacom AI Lab, and How was It Developed?

Softacom AI Lab is the result of Softacom’s hands-on experience across the entire lifecycle of computer vision solutions. From business analysis to integrating trained models into a client’s product, we’ve mastered every step of the process.

This includes:

1. Analyzing business needs
2. Assessing feasibility and accuracy
3. Selecting the right AI architecture
4. Collecting and preparing datasets
5. Generating additional insights
6. Training model
7. Validating results
8. Integrating them into a product
9. Testing, and
10. Releasing the solution

Throughout this journey, the Softacom team often had to rely on various third-party tools (open-source or commercial), while overcoming integration challenges, automating workflows, and addressing functional gaps.

Our ultimate goal was to make the pipeline clear and accessible, even for non-developers, so they wouldn’t need to code in Python or navigate complex consoles.

This is how Softacom AI Lab began to take shape—a solution designed to save time for both us and our clients by accelerating AI implementation.

Time is critical in today’s fast-paced world. Delays in deploying AI solutions risk falling behind rapidly evolving technologies and standards, which can render even groundbreaking innovations obsolete.

In short, Softacom AI Lab is software for Windows and Linux that allows users—whether our team or clients with our guidance—to prepare data, train models, and visualize results. This way, the clients can assess whether a trained neural network model is ready for commercial use or if further iteration is needed.

The project evolved organically, driven by real-world needs rather than being built in isolation. Initially, it served as an internal tool for visualization and validation, with new features added over time.

Eventually, it matured into what is now officially known as Softacom AI Lab.

What Business Challenges Does Softacom Plans to Address with AI?

We believe in making the world better—through actions, ideas, and even commercial products. Businesses thrive by solving customer needs, driving sales, and optimizing internal operations to remain profitable. This is where we come in.

Currently, our focus is on B2B solutions, assisting clients in two key areas:

1. Developing new AI-powered features for their products.
2. Implementing internal AI solutions to reduce costs, improve efficiency, and boost profitability.

The specific challenges vary widely depending on the industry. Whether it’s security, medical equipment, or agriculture, AI—particularly computer vision—offers transformative potential. Computer vision goes beyond simply “seeing”; it enables understanding, analysis, decision-making, and continuous learning from new situations.

At Softacom, we aim to unlock this potential, tailoring AI solutions to solve real-world business problems and drive innovation.

What Sets Softacom’s AI Solutions Apart

Softacom offers a three-tiered advantage:

1. Proven AI Architecture: We leverage a comprehensive toolkit of components developed by mathematicians, researchers, and various companies. These tools are combined and optimized like building blocks. We’ve already validated what works and identified the best combinations for specific challenges.
2. Softacom AI Lab: Our proprietary software facilitates the management, testing, and understanding of AI architectures. We deliver real, tangible solutions—not just concepts.
3. Custom Optimization Algorithms: Our post-analysis algorithms enhance neural network outputs. They optimize input data, streamline processing, minimize computational load, and improve result accuracy through additional checks. These innovations accelerate AI adoption for our clients.

Importantly, our AI architecture supports both on-premises and cloud deployments, addressing diverse business needs.

Technologies Behind Softacom AI Solutions

To deliver effective solutions, the Softacom team relies on a wide range of development tools and technologies.

Specifically for AI and neural networks, our stack includes:

• Languages and libraries: Python (e.g., TensorFlow, Keras) and C# (e.g., ML.NET)
• Tools: Softacom AI Lab, alongside third-party utilities for data labeling and dataset preparation
• Pre-trained models: Both open-source and commercial options

Beyond AI, we work with various tools and languages, including Visual Studio, RAD Studio/Object Pascal, Python, C#, and C++.

Our secret ingredient? A unique blend of expertise, innovation, and dedication.

Value of Softacom AI Solutions for Businesses

Softacom AI solutions contribute directly to our clients’ products, becoming an integral part of their businesses. By helping our Clients innovate and optimize, we enhance their ability to serve their customers even more effectively.

For end-users of these products, AI solutions deliver tailored value based on the industry. For example:

• Security: Improved safety measures
• Healthcare: Enhanced patient outcomes
• Agriculture and sustainability: Promoting sustainable practices

We create tools that transform industries by solving real problems with AI. You are welcome to learn more from the dedicated page on the AI transformation services.

Can You Share a Success Story?

Although this is a predictable and popular question, it still caught me off guard because I’d rather not give a standard example. Here’s one:

We were approached by a large security company that offers x-ray scanning systems for detecting prohibited items, similar to those used at airport security checks.

Naturally, these systems come with software from the manufacturer, which works well, including using trained neural network models to automatically detect items prohibited for air travel (sharp objects, weapons, etc.) and assist operators in identifying them.

Let’s call this Company A.

Company A was participating in a tender to supply these devices to an organization, which we’ll call Organization B. This organization controls access to its premises and monitors what items employees and visitors can bring in and out. The types of items didn’t fit the standard set of items the system was designed for, so the system couldn’t help the operator detect items quickly. Additionally, the list of items and their characteristics changed periodically.

To address this, Company A reached out to us, and together with the equipment manufacturer, we developed a system that could process video and images through additional neural network models and generate events in the manufacturer’s standard interface. We implemented the entire process of creating and training neural networks to meet Organization B’s needs, trained Company A and Organization B on how to fine-tune models for new object types, connected our new system to the equipment’s SDK, and carried out testing.

As a result, Company A won the multi-million-dollar tender, and Organization B received the required modern AI-based functionality, improving security, reducing financial losses.

Which Industries Can Benefit Most from AI Solutions?

From both theoretical and practical experience, I can say it depends on what I know and what we’ve implemented successfully.

Here’s my perspective:

Rather than business directions, I would focus on industries—specifically, security, healthcare, and various office and corporate business solutions.

And I’d like to highlight the latter—we don’t just work with cameras and photos to analyze real-world data; we can also analyze information from monitors, documents, archives, etc.

We’re there where a human eye and brain are needed for information analysis.

AI systems trained for specific tasks can analyze human behavior, assess work environments, evaluate efficiency, and replace routine tasks. The possibilities are vast.

For example:

An AI system could analyze how someone works with a particular program and provide statistics—not from within the program itself (the classic approach), but from an external AI system trained to analyze work in SAP, email management, and more. This type of generic system can be further trained or adapted to new tasks.

How Soon Can Businesses See Results from Implementing AI Solutions

The timeline depends on the implementation cycle of new functionality in our clients’ products. 

As soon as they begin selling their new or improved solutions, they’ll start seeing financial returns.

However, to get there, we must go through the entire process together, which includes business analysis, selection and adaptation of our AI architecture, data preparation, neural network training, result validation, integration into existing software or new development, testing, and release.

An important point I haven’t mentioned yet is the equipment and computational resources required to train neural networks and perform visual data analysis. Training neural systems demands modern equipment, which directly impacts how quickly we can experiment, test, and validate results.

Sometimes clients already have the necessary equipment, other times they need to purchase it, and sometimes we handle training on our own equipment or in the cloud (e.g., AWS, Google). 

For analyzing real data in client products, end-user equipment will be used, which can vary depending on the specific implementation—whether it’s a computer, cloud system, single-board computer in a mobility device, or even on a drone.

Instead of Summary: What’s Next for AI?

This is a topic that could spark long discussions.

It’s a global trend, with entire conferences dedicated to professionals speculating about what’s ahead. 

I believe this is not a temporary trend, like Blockchain, which many once thought would be a fundamental part of all software in the future. But AI is different—it’s not just one thing; it’s a collection of technologies from various fields, industries, and development tools that touch every aspect of our existence and life.

AI will evolve in all directions toward the development of General AI—until humanity creates something similar to itself.

Yes, it sounds utopian, but on a more grounded note, I see promising developments where AI will optimize and enhance human productivity, perform tasks that humans can’t. We may soon see virtual AI sales managers selling products or services to virtual AI employees in other companies, while AI marketers optimize company information for AI buyers.

Apologies if I went a bit speculative and futuristic rather than sticking to predictions.
We thank Serge Pilko for sharing his valuable insights on the topic. Should you have any further questions about Softacom’s AI transformation services, business ideas, or feedback, please don’t hesitate to reach out to us. We welcome your thoughts and look forward to connecting.