This ongoing project focuses on the design and development of an integrated enterprise IT system that connects ERP, internal and external communication, web applications, and analytical tools into a single, coherent platform. The solution follows an AI-first approach, treating artificial intelligence as a core element of the system architecture.

Problem:
Rapidly changing business requirements and the fast evolution of AI technologies have made traditional ERP systems, even when supported by standard office tools, insufficient to ensure operational scalability. Organizations need a more holistic approach to information systems to improve business processes, information flow, and highly personalized customer experiences, while reducing overall operational costs and employee workload.

Solution:
The project defines a unified set of tools, software components, and practices operating as one integrated system. The architecture is based on microservices and event-driven principles, with automated workflows supervised and enhanced by AI. This AI-first design enables flexibility, scalability, and continuous improvement.

Conceptual example (communication):
In the target architecture, company communication can be analyzed by AI to support continuous information flow. Documents may be automatically classified, with key data synchronized with ERP. Communication actors can be grouped into behavioral patterns and customer personas, which can be used for web experience personalization or for generating high-quality automated responses. The system can generate high-level events such as “conversation updated”, “customer sentiment changed”, or “new sales opportunity”, while providing management with advanced analytics via natural language interfaces. Appropriate reports are automatically delivered to your inbox.

Result:
Improved business process quality, reduced operational costs, and a scalable foundation for long-term, AI-driven growth.

The project focused on the design and implementation of an automated system for document circulation, classification, and storage within an enterprise environment. Its objective was to reduce manual workload, minimize errors, and improve data availability and quality.

The key challenge was managing a large and growing volume of documents requiring classification, storage, replication of information, and efficient retrieval. Manual processing generated high operational costs and increased the risk of inconsistencies, while delayed or missing document handling could lead to financial and compliance issues.

The solution was an integrated platform capable of ingesting documents from multiple sources (e.g. email inboxes, scanners) and automatically classifying, indexing, and extracting structured data from unstructured content. Extracted information was directly transferred to downstream systems, such as ERP (e.g. purchase invoices). The system also provided fast and flexible document search, including queries based on partial or incomplete content.

Key challenges included high variability of document formats, inconsistent standards, complex and multi-page documents, low-quality scans, and security requirements such as fraud detection (e.g. fake invoice scams).

As a result, the implemented system significantly improved document management efficiency, reduced administrative workload, increased data quality, and eliminated errors related to delayed or missing document processing. Advanced search capabilities improved organizational transparency and operational awareness.

Conclusions:
Intelligent document automation enables substantial cost reduction, improves data reliability, and strengthens integration between document workflows and core business systems while ensuring scalability and security.

Machine Learning–Based Optimization of Energy Storage and Resale Strategy for Residential Photovoltaic SystemsMachine Learning–Based Optimization of Energy Storage and Resale Strategy for Residential Photovoltaic Systems

Research project focused on designing machine learning–driven algorithms for optimizing short-term (24-hour horizon) energy storage and resale strategies in residential photovoltaic systems equipped with battery storage.

Problem
Residential photovoltaic installations operate under multiple constraints: limited and time-dependent energy production, finite battery storage capacity, variable household energy consumption, and fluctuating electricity resale prices. Poor short-term planning may lead either to premature energy sales resulting in shortages for self-consumption, or excessive storage that misses optimal market prices—both significantly reducing economic efficiency and return on investment.

Approach
The system integrated multiple external data sources via APIs, including historical energy consumption, photovoltaic production, electricity market prices, and weather forecasts.
Machine learning models were trained to predict production, consumption, and price dynamics over the next 24 hours. Based on these forecasts, the system generated an optimal daily strategy defining when energy should be stored in batteries and when it should be sold back to the grid, while ensuring sufficient availability for household self-consumption.
To improve decision quality and robustness, the solution combined several ML models with a genetic algorithm responsible for dynamically optimizing model weights.

Result
The project delivered a practical decision-support tool enabling intelligent, data-driven energy storage and resale planning. The solution significantly improved the economic performance of residential photovoltaic installations while maintaining energy security for household consumption.

The project involved a system handling over 15 million requests per day, originally built as a monolithic application. The architecture suffered from multiple bottlenecks, limited scalability, and long downtime periods, which resulted in high operational costs and increased business risk. The situation was further complicated by the fact that part of the system delivered security-critical services, where any downtime could lead to potentially very high financial and operational impact.

My responsibility was to design and implement a new system architecture. I led the transformation into a distributed, highly redundant system with a clear separation of responsibilities across individual nodes. The solution was based on multiple independent microservices deployed on VPS infrastructure, enabling flexible resource management and independent scaling of system components.

Project outcomes:

• horizontally scalable architecture ready for further traffic growth,
• high availability for critical services, achieving 99.99% uptime (SLA) in production,
• significantly reduced system response times,
• measurable reduction in operational costs,
• elimination of single points of failure and increased system resilience.

This project had a direct impact on business continuity, operational security, and the long-term scalability of the platform.

The system was based on a custom-built web application supporting sales processes, integrated with a geographically remote ERP system. The architecture caused significant data retrieval latency and instability during periods of increased user traffic.

Problem

• High communication latency with the ERP system
• ERP overload during peak traffic generated by the sales application
• Risk of web application downtime and degraded end-user experience

Constraints

• No possibility to modify the ERP system (technical debt)
• No option to change the geographical location of system components
• Limited ability to modify the web application code

Solution
Design and implementation of a dedicated proxy layer, deployed close to the web application and acting as a buffering and optimization layer:

Phase 1: collection of detailed metrics and operational data to identify real sources of latency and to design effective optimization mechanisms

Phase 2: implementation of aggressive caching strategies (synchronous and asynchronous), local data manipulation, and a significant reduction of direct calls to the ERP system

Results

• Up to 90% reduction in ERP-generated latency
• Complete elimination of ERP saturation issues
• Stabilization of the sales web application and removal of downtime caused by overloads
• Significant improvement in system performance and scalability without violating existing technological constraints

The project demonstrates a practical approach to solving performance-related problems in environments with significant technological and organizational constraints.

This project delivered a full redesign and virtualization of infrastructure for a hosting provider managing over 1,000 active domains.

Challenge
The legacy environment relied heavily on physical servers, generating high operational costs, technical debt, and strong dependence on hardware stability. Hardware failures directly impacted service availability, while scaling and configuration changes required extensive manual effort and long implementation times.

Solution
A new, fully virtualized infrastructure was designed and deployed. The architecture was based on smaller, role-oriented VPS nodes, supported by extensive automation. This approach separated services from physical hardware, significantly improving scalability, resilience, and operational flexibility.

Scope of Work
The project included selecting and configuring physical servers with dedicated compute and storage roles, deploying virtualization and distributed storage technologies, implementing automation, orchestration, and monitoring systems, and migrating all existing hosting services to the new environment.

Migration Challenge
A critical requirement was to minimize customer impact during migration. This was achieved through advanced migration tools and custom automation, enabling controlled, phased transitions without service interruptions.

Results & Impact
The new platform improved security, reduced operational costs, and isolated hardware failures from hosted services. Automation accelerated provisioning and configuration changes, while the modular design enabled horizontal scaling and future expansion.

Key Takeaway
The project transformed a hardware-dependent hosting platform into a scalable, resilient, and automation-driven infrastructure, improving reliability, efficiency, and long-term growth potential.