Lead AI engineer and solution architect with 23+ years in software and a decade in machine learning. I design and ship agentic LLM / RAG systems, computer vision and predictive-maintenance solutions, owning them from architecture to production. Based in Cluj-Napoca, Romania.
Machine LearningDeep Learning
Machine Learning Engineer with over 20 years of software engineering experience specializing in deep learning and advanced machine learning techniques. Highly skilled in software architecture and software engineering. Demonstrates a strong commitment to researching state-of-the-art machine learning solutions to address real-world challenges. Proficient in developing AI-enabled products that enhance operational efficiency and innovation.
Experience
Professional experience10+ years
Freelancer since2016
Languages
EnglishFluent
Project history
AI Solution Architect / Team Lead – E.ON MyLine Conversational AI
RebelDot Solutions (client: E.ON)
Energy, Water & Environment
Customer-facing conversational assistant integrated into the E.ON MyLine customer-care app, letting users manage contracts, invoices, payment history and general inquiries. Rolled out to millions of customers; running on Azure AKS with the knowledge platform on Azure Databricks.
Solution architect: designed the solution that won the proposal and defined the overall architecture (conversational layer, AI pipelines, backend integrations, cloud deployment); led the team from design through production.
Led development of the knowledge base used by the RAG pipelines: ingestion, transformation and organisation of structured and unstructured content in Azure Databricks, exposed to LangGraph-orchestrated agents.
Designed the AI pipelines (retrieval, LLM orchestration, tool calls into contract / invoice / payment systems) for scalability and reliability at high volume.
Tech: Microsoft Azure (AKS), Azure Databricks, Python, FastAPI, LangGraph, Langfuse
Lead AI Engineer / Architect – GEA CoPilot (Industrial RAG Assistant)
RebelDot Solutions (client: GEA)
Industry & Mechanical Engineering
AI-powered support assistant helping technicians diagnose and resolve complex issues with industrial equipment: an agentic RAG conversational interface over internal knowledge bases and enterprise systems, with the data and AI platform on Azure Databricks.
Team lead and architect: drove technical direction and owned decisions from design through production.
Designed document-ingestion pipelines as Databricks Jobs (medallion bronze / silver / gold Delta tables) feeding a Vector Search index that grounds the RAG pipelines.
Terraform-managed Databricks workspaces with release pipelines promoting tables, indexes and models from dev to staging and production.
Fine-tuned cross-encoder rerankers on user-feedback data, tracked in MLflow and registered in Unity Catalog.
Built the Python backend and agentic AI pipelines and led integration with GEA internal systems.
AI Solution Architect – Rebel FLUX (AI Workflow Automation)
RebelDot Solutions
Internet & IT
Collaborative meta-harness platform that orchestrates agentic, LLM-driven SDLC workflows, letting teams model, customize and execute autonomous-agent workflows while keeping human oversight at critical decision points.
Led the exploratory, AI-native architecture definition.
Delivered the first MVP using agentic coding harnesses.
AI Engineer (Freelance) – Automated Image-Editing Pipeline
AND Digital
Internet & IT
Short-term freelance engagement building an automated image-editing pipeline for a major European real-estate company, using instruction-based (text-to-image) editing to scale listing-image enhancement while ensuring consistent quality.
Designed and implemented end-to-end image-editing pipelines.
Experimented with instruction-based image-editing models across edit scenarios and built workflows for large-scale batch processing.
Senior ML Engineer – DINAMO (EU-funded R&D, Robotics)
NTT DATA Romania
Industry & Mechanical Engineering
EU-funded, voice-enabled robot-control system enabling natural-language interaction (e.g. “follow me”), with an automated pipeline for generating synthetic training data and building robust intent-recognition models for real-world robotic voice control.
Authored the technical proposal that secured EU funding.
Designed and implemented an end-to-end synthetic-data-generation pipeline using LLMs.
Developed intent-recognition and named-entity-recognition models.
Senior ML Engineer – AI-PRISM (EU-funded R&D, Robotics)
NTT DATA Romania
Industry & Mechanical Engineering
EU-funded research initiative advancing human-robot collaboration, coordinated by the Technical University of Valencia: a consortium of 3 universities, 3 research institutions and 20 companies working on ambient sensing, perception, command & control and safety.
Created datasets for human-activity recognition from video.
Trained and deployed highly accurate human-activity-recognition models (SlowFast, ViViT).
Tech: Deep Learning, CNNs, Vision Transformers, Python, PyTorch, Torchvision, Continuous ML (CML), GitHub Actions, Docker, Docker Compose
Lead ML Engineer – Glas.AI (Automotive AI)
DevTel & ThirdSpace Auto & Glas.AI
Automotive & Vehicle Manufacturing
Customizable pipeline for building conversational assistants focused on offline natural-language understanding for in-car environments, with real-time, on-device speech and intent recognition for low-latency, privacy-preserving voice interactions.
Developed incremental speech-to-text models with a focus on real-time inference on mobile devices.
Led the team developing intent-recognition and named-entity-recognition models.
Tech: Speech-to-text (ASR), RNNs, Python, TensorFlow, Keras, TensorFlow Lite
Lead ML Engineer – Vibrat.IO (Predictive Maintenance)
DevTel & ThirdSpace Auto & Glas.AI
Industry & Mechanical Engineering
End-to-end condition-monitoring platform for predictive maintenance via vibration-signal analysis. Grew from R&D into a production-grade solution deployed across industrial environments, including a bakery and a wind farm.
Led the R&D team developing predictive models for fault detection (classification / anomaly detection) and Remaining Useful Life (RUL) estimation.
Built simulators to generate controlled training and validation data, and developed the application layer for interaction and monitoring.
Tech: Time-series / signal processing, Deep Learning, CNNs, Python, NumPy, Pandas, PyTorch, .NET, ASP.NET MVC, Blazor
Lead ML Engineer – PiceaOne (Retail Computer Vision)
DevTel & ThirdSpace Auto & Glas.AI
Goods & Retail
R&D project recognizing product stickers and identifying mobile-phone models, defects and overall condition from live video streams, optimized for mobile deployment.
Led the team developing models for object detection, text segmentation, OCR and information extraction.
Created custom object-detection datasets and trained highly efficient, accurate models.
Implemented a then-novel approach to information extraction from text using neighborhood encoding.
Short-term freelance R&D engagement designing a recommendation system for BringATrailer.com from years of user-interaction data, to improve item discovery and engagement in an auction marketplace.
Performed extensive exploratory analysis on large-scale interaction data.
Designed and evaluated sequence-based (LSTM), collaborative-filtering and content-based approaches; engineered interaction signals with custom weight normalization.
Implemented and validated a matrix-factorization approach using LightFM.
Automatic Invoice Processing (enterprise / accounts-payable automation): R&D for automatic data extraction from invoices at scale, handling documents with highly diverse layouts and visual structures.
Modernization migrating a legacy enterprise CRM from a Progress-based system to a modern WPF desktop application.
Designed a reusable WPF component library supporting application look-and-feel and workflows.
Developed an API layer for remote-procedure calling of Progress stored procedures.
Tech: WPF, C#
Senior Software Engineer & Solution Architect – IoC Middleware (Healthcare)
Yonder
Pharmaceuticals & Medical Technology
Long-term development of a cross-platform inversion-of-control (IoC) middleware that became the core application infrastructure for six production healthcare products, enabling large-scale modernization of legacy systems to WPF and single-page applications with service-component backends.
Designed and implemented a custom dependency-injection container with features and performance on par with or exceeding Google Guice, ported to C#, Java and JavaScript.
Led the application-infrastructure implementation and supported six production healthcare applications built on the platform.
Tech: C#, WPF, Java, Dependency Injection / IoC patterns
Software Engineer – ArchiCAD Next-Generation R&D
Yonder (client: Graphisoft)
Architecture & Civil Engineering
R&D for a next-generation version of ArchiCAD, a leading CAD application used in architecture and building design.
Worked on core features such as solid-modeling components for geometric representation and manipulation.
Integrated and used 3D rendering capabilities via OpenSceneGraph.
System-level components for Pumatech Intellisync, integrating personal-information managers (PIMs) with enterprise messaging systems such as Lotus Notes.
Developed C++ components for data synchronization between Intellisync and Lotus Notes.