Artificial Intelligence

• SNN

• Brain, a platform for creating energy consumption and generation prediction models using DL architectures based on convolutional networks + LSTM.

• Cicero, systematic research of DL models for malware classification using features extracted from memory dumps. Application of XAI (Explainable AI) to understand the features most relevant to the models

• Design and selection of tools across all processing layers.
• Selection and use of the most appropriate databases (SQL and NoSQL: Cassandra, Postgres, FHIR, Azure Data Lake, etc.) for cloud or on-premise environments.
• Development of client-server systems and communication tools (Kafka, MQTT).
• Tools and programming languages: Jupyter, R Studio, Python, R, Java, C, C++, C#.
• Applications: Data collection from multiple sensors, decision-making across multiple devices and levels.

• Definition and design of systems tailored to specific requirements and specifications.
• Development of custom algorithms for specific use conditions.
• Web deployment of constraint-based schedulers allowing user interaction and real-time changes.
• Integration with multiple databases.
• Applications: Scheduling of medical appointments, process optimization under constraints like size, staff availability, etc.

• Selection of the target prediction variables.
• Identification of the most suitable time series algorithm and the most accurate forecasting periods.
• Development and deployment of the algorithms in real-world environments.
• Applications: Production and stock forecasting in factories. Pattern detection in online purchasing behavior.

• RPA systems for process automation and web scraping using Python and R.
• XPath identification and use of UIPath.
• Applications: Web scraping for administration, insurance, external websites, e-commerce, etc.

• Design of predictive algorithms deployed at the edge in industrial environments and industrial PLC devices.
• Development of custom algorithms tailored to specific usage conditions.
• Algorithm development on Beckhoff PLC systems using Beckhoff ML and Beckhoff Vision tools.
• Data transmission and report generation.
• Applications: Anomalous pattern control systems in machines, enhancing tool protection in production plants.

• Definition, design, and implementation of a GPU-accelerated computing system in rack configuration, with operational standards and scalability aligned with data center-level infrastructure.
• Advice on hardware components and platforms for project deployment across different sectors.
• Integration with interactive development environments such as JupyterLab and Visual Studio Code, allowing seamless workflows for development, training, and GPU-based model deployment.
• Deployment of Docker containers with preconfigured environments for testing.
• Applications: GPU rack systems accessible to multiple models.

Creation, configuration, customization, and deployment of conversational agents using written language, powered by LLM models. This AI technology enables assistance with specific tasks and queries in natural language. We focus on improving the quality, fluency, and factual accuracy of the chatbot’s responses. It is always advisable to implement “guardrails” to ensure the ChatBot only engages in conversations related to its intended topics and objectives.

Like ChatBots, these agents assist users in a conversational manner. Unlike ChatBots, which use written text to communicate, SpeechBots use spoken language. They are capable of understanding user voice input and responding with synthetic speech. This is achieved through specific modules that convert voice to text and text to speech. It’s important to optimize these conversational flows to ensure smooth, frictionless interaction with the SpeechBot.

These LLMs or other generative AI models can be accessed via API from various providers. Companies like OpenAI, Anthropic, or Google offer their models for specific applications on a pay-per-use basis. However, at ITCL we also invest in Open Source GenAI models like LLAMA, DeepSeek, or Gemma. We are researching a concept we call the LLM-Provider, which enables dynamic and customized use of these Open Source models with GPU resources—ensuring privacy and confidentiality by design.

One of the most impressive applications of AI Agents is the generation and execution of programming code. This task, which requires significant reasoning ability from humans, is ideal for automation through AI Agents because it is verifiable. A code snippet can be executed in an environment to verify its correctness. The AI Agent can not only write the code, but also create the execution environment, install necessary dependencies, and deploy the service. This enables non-programmers to build their own coding projects or support prototype development.

AI Agents are highly effective tools for quickly and directly accessing specific information from vast data sources. Tools such as RAG (Retrieval Augmented Generation), KAG (Knowledge Augmented Generation), Web-Search, or File-Search enable us to design AI Agents that can retrieve specific data swiftly and automatically, presenting it to the user in the desired format.

In many companies, there are departments focused on reporting or document generation. Often, these documents are compiled from scattered sources, requiring manual gathering, merging, and formatting. When this type of task is identified, GenAI Agents for document generation become a powerful tool for automation, delivering synthesized information in a finalized document.

At ITCL, we work to deploy the models powering these AI Agents not only via APIs in central servers, but also on edge devices. These devices—such as NVIDIA’s Jetson family or Xilinx FPGAs—are resource-constrained and cost-effective, posing a challenge for implementing large models like LLMs. However, once this challenge is overcome, they allow reduced latency and localized systems with greater privacy for data processing.

At our AI Agents R&D Unit, we not only build AI Agents and Generative AI applications to automate specific tasks, but also stay up to date with AI tool integration to improve daily operations—for ourselves and our clients. If you’re unsure how AI can enhance and integrate into your company, feel free to reach out and talk to us.