I am an Engineer and Practitioner operating at the intersection of physical structures, advanced data science, and AI/ML engineering. I specialize in delivering end-to-end ML/AI systems — bridging the gap between deep mathematical research, model training and fine-tuning, and robust software deployment.

With a background in structural analysis in engineering, I bring a unique domain-first approach to data. I don’t just apply algorithms; I understand the physical, spatial, and structural nature of the data I work with.

I build systems that are mathematically sound, containerized for the cloud, and deeply aligned with real-world engineering and business constraints.

Founder & Principal Architect · Wrocław, Poland · since Jan. 2026

Manifold Labs is a deep-tech engineering venture transforming Structural Health Monitoring (SHM) for critical infrastructure (dams, bridges, tunnels) by combining Topological Data Analysis (TDA) with fault-tolerant edge telemetry. Deployed purely as a non-invasive software overlay on top of existing SCADA/cloud repositories, the platform detects subtle, nonlinear failure mechanisms before physical thresholds are breached.

What I design and build

• Dual-Track TDA — core analytical architecture using persistent homology (H₀, H₁) and Banach space projections (Lᵖ persistence landscapes) to detect dynamical bifurcations and Critical Slowing Down in geotechnical media.
• Cross-Verification Engine — deterministic decision engine decoupling sensor degradation and cable noise (Track A: Sensor Health) from real pore-pressure anomalies (Track B: Structural Behavior), slashing false positive rates.
• Heterogeneous Sensor Fusion — mathematical foundation for multi-sensor state-space analysis, integrating piezometers, inclinometers, and drainage seepage data into a unified, sensor-agnostic geometry.
• End-to-End System Architecture — core Python mathematical engine integrated with fault-tolerant, telecommunications-grade Elixir/BEAM edge telemetry.

Focus: Topological Data Analysis · Structural Health Monitoring · Geotechnical Engineering · Early Warning Systems · Critical Infrastructure Safety

• Production-Grade Analytics — developing and refactoring high-performance codebases in Python and R, with a strong focus on clean architecture, unit testing, and long-term maintainability.
• Systems Architecture — modular, scalable API-based services and analytical engines. CI/CD pipelines, containerization (Docker/Podman), and code stability in complex, hybrid environments.
• Domain-Driven Engineering — translating complex mathematical and statistical requirements into production-ready software components.
• Engineering Standards — clean, modular code, reproducible research workflows, and automated testing across cross-disciplinary teams.

• Topological Feature Extraction (Python / giotto-tda): pipeline extracting persistence diagrams from multidimensional data for robust anomaly and structural state detection.
• Cross-Technology Communication (Elixir / Python): integration of the Python analysis engine with a high-performance Elixir backend for real-time data streams and immediate anomaly flags.
• Scientific Reporting (Quarto): automated reporting system generating dynamic, interactive documents with topological visualizations.
• Domain-Driven Feature Engineering: mapping abstract topological changes (e.g. persistence entropy) to physical anomalies and system degradation.

• Fine-tuned a Donut (VDU) Transformer via Hugging Face for OCR-free entity extraction.
• Custom Python annotation pipeline for a specialized dataset of Polish receipts with complex spatial layouts.
• OOP-based PyTorch inference module (preprocessing, tensor ops, custom token decoding) plus a training orchestrator.
• Containerized inference with Docker; git-based versioning over interactive notebooks (Emacs org-babel).

• Parsed complex financial PDFs (balance sheets, P&L) into structured Markdown.
• Conversational interface over OpenAI models for querying report data.
• Structured, reliable extraction via Instructor and Pydantic schemas.

• Combined classical ML (scikit-learn) with LLMs (OpenAI) to interpret model outputs.
• Feature extraction and normalization from textual data using Pydantic.
• LLM-generated technical explanations for tabular model predictions (XAI).

• Multi-agent system (OpenAI API, Instructor) for drafting and validating contracts against the Civil Code.
• Pydantic-based structural validation of inputs and LLM outputs.
• Python (OOP), RAG, CoAct.

• Transfer Learning on DenseNet121 (TensorFlow/Keras) with custom dense classification heads.
• Custom learning-rate schedules and early stopping, prioritizing high recall to minimize critical false negatives.
• Class-imbalance mitigation via spatial and pixel-level augmentations (Albumentations) and stratified splits.
• Reproducibility: Docker, pyproject.toml, Jupytext-versioned notebooks.

• Postgraduate, Data Science — Warsaw University of Technology
• Postgraduate, Geotechnical Engineering — WULS-SGGW
• M.Eng., Hydraulic Engineering — Wrocław University of Environmental and Life Sciences
• Workshop, Practical LLM — DataWorkshop.eu
• Workshop, Machine Learning — DataWorkshop.eu

• Polish — C2
• English — B2/C1

• 🌐 manifoldlabs.eu
• ✉️ contact@manifoldlabs.eu