{ "basics": { "name": "Leo Vaicer", "label": "Systems & Data Engineer", "summary": "Systems & Data Engineer, Applied Mathematics background. Seven years on one recurring problem: recovering reliable state from unreliable measurement, and keeping that recovery running under production load. Wind resource assessment from raw SODAR and met-mast signal; clinical data platforms turning unstructured hospital records into verifiable schema; adaptive control for high-altitude flight under non-linear turbulence. I build the model and the infrastructure it runs on, because they're the same problem.", "location": "Calgary, Canada", "url": "https://www.zeelex.me/", "profiles": [ { "network": "LinkedIn", "username": "", "url": "https://www.linkedin.com/in/zeelexes" } ] }, "work": [ { "name": "GE Vernova", "position": "Software Developer", "location": "Barcelona, Spain", "start_date": "Aug 2024", "end_date": "Dec 2025", "summary": "", "highlights": [ "Built and maintained the Python ecosystem behind wind resource assessment - SODAR (Sonic Detection and Ranging) and met-mast processing, vectorized in NumPy.", "Implemented detrending and temporal decomposition to analyze air-mass density stability and hub/mast positioning, feeding turbine siting decisions.", "Rewrote critical Airflow DAGs around vectorized algorithms: 5–10X faster task execution and far fewer pipeline failures.", "Migrated 4+ core services to Python 3.x, writing LLM-based static analysis to automate type-hinting, signature refactoring, and dependency resolution (UV/Ruff), and built pymolt to automate behavioral verification.", "Built the observability stack - Grafana, Loki, Prometheus — over GPU-accelerated Kubernetes clusters, cutting incident resolution time through real-time tracking." ] }, { "name": "Trialing Health S.L.", "position": "Data Engineer", "location": "Barcelona, Spain", "start_date": "Dec 2023", "end_date": "Aug 2024", "summary": "", "highlights": [ "Set up an AWS S3 data lake and rebuilt the Airflow ETL around it, cutting data-to-insight latency from 120 minutes to 40.", "Designed a multi-tier validation framework with a dedicated quarantine zone for manual verification, holding data integrity across engineering and data-management teams.", "Built an ingestion engine (Regex, Pydantic) that turned non-standardized scientific records into a high-fidelity schema, letting physicians track drug efficacy and treatment outcomes in real time.", "Deployed a unified medical data platform aggregating unstructured research archives from hundreds of hospitals and clinical centers across Spain and Portugal." ] }, { "name": "Lemay.ai", "position": "DevOps and MLOps", "location": "Toronto, Canada", "start_date": "Dec 2022", "end_date": "May 2023", "summary": "", "highlights": [ "Administered bare-metal Proxmox hypervisors (LVM partition resizing, hardware allocation) alongside core AWS cloud infrastructure (EKS, EC2, S3).", "Implemented Apache Kafka & Kafka Streams for microservice event streaming, establishing baseline observability across event lifecycles.", "Designed a secure data ingestion layer enforcing strict Role-Based Access Control (RBAC) across dataset onboarding.", "Deployed self-hosted GitLab CI/CD pipelines and managed version-controlled Nginx reverse-proxy configurations." ] }, { "name": "Contract (DataArt, Inmost, Virtuace)", "position": "Data & MLOps Engineer", "location": "", "start_date": "Jul 2021", "end_date": "Aug 2023", "summary": "", "highlights": [ "Delivered infrastructure and ML work for international clients on cloud migration and high-load systems: disaster recovery on AWS to 99% availability, and 15–25% lower storage and processing costs through S3/Lambda resource optimization.", "Decoupled monolithic ML pipelines into modular microservices using MLflow, and integrated DVC (Data Version Control) for reproducible dataset and model tracking.", "Led on-premise to cloud migrations, optimized PostgreSQL aggregation schemas, and standardized CI/CD across client engineering teams." ] }, { "name": "Sembly AI — full-time, concurrent with BSc studies", "position": "Backend Engineer, ML Infrastructure", "location": "Odesa, Ukraine", "start_date": "Oct 2019", "end_date": "Jul 2021", "summary": "", "highlights": [ "Architected a high-throughput Kafka audio ingestion system for real-time stream processing, driving a 30% improvement in system responsiveness.", "Decoupled monolithic AI platform into fault-tolerant microservices and overhauled the NLP inference path — slashing deployment latency by 40% and boosting inference speed by 25%.", "Established end-to-end automated model training and validation workflows, cutting model time-to-production by 50%." ] }, { "name": "Elliscope (co-founded) - part-time, concurrent with Sembly AI", "position": "R&D Engineer, Control & Energy Systems", "location": "", "start_date": "Jul 2019", "end_date": "Mar 2022", "summary": "R&D venture on autonomous high-altitude airships and hybrid energy propulsion. Presented at WebSummit 2021. https://elliscope.zeelex.me", "highlights": [ "Developed adaptive control algorithms - Kalman filtering and Model Predictive Control - compensating for non-linear turbulence and buoyancy loss at 10–12 km, and assessed controllability across atmospheric profiles.", "Modeled hydrogen fuel-cell integration and energy balance for a theoretical 5+ week flight endurance, identifying the critical lift-to-weight constraints for zero-emission propulsion.", "Designed and tested scale-model demonstrators validating stability and sensor-integration hypotheses; published findings within the Ukrainian Space Association." ] } ], "education": [ { "institution": "Universitat Autònoma de Barcelona (UAB)", "study_type": "Graduate coursework — MSc in Modeling for Science and Engineering", "start_date": "2025", "end_date": "2026", "url": "", "highlights": [ "Mathematical modeling, dynamical systems and complexity, high-performance computing (HPC)" ] }, { "institution": "Odesa I.I.Mechnikov National University", "study_type": "BSc, Applied Mathematics and Computing", "start_date": "2019", "end_date": "2023", "url": "", "highlights": [ "Numerical methods, optimization and control theory, machine learning, econometrics", "Completed while working full-time" ] } ], "skills": [ { "name": "Programming Languages", "keywords": [ "Expert: Python", "Competent: C/C++", "Rust", "Java" ] }, { "name": "Mathematical Modeling & HPC", "keywords": [ "Numerical Methods", "Optimization & Control Theory", "High-Performance Computing (HPC)", "Time-Series Analysis", "Vectorized Algorithms", "Dynamical Systems" ] }, { "name": "State Estimation & Signal Processing", "keywords": [ "Kalman filtering", "Model Predictive Control (MPC)", "time-series analysis & decomposition", "detrending", "signal processing", "control theory", "data validation architecture" ] }, { "name": "Data Engineering", "keywords": [ "Apache Airflow", "Apache Spark", "Apache Kafka", "Celery", "dbt", "ETL/ELT", "data lakes (Iceberg/Delta)", "data modeling", "data quality", "distributed systems" ] }, { "name": "Cloud & Infrastructure", "keywords": [ "AWS (S3, Lambda, EC2, EKS, Glue)", "Kubernetes", "Docker", "Terraform", "Helm", "CI/CD (GitHub Actions, GitLab CI)", "Linux", "microservices", "infrastructure as code" ] }, { "name": "ML & AI", "keywords": [ "PyTorch", "TensorFlow", "scikit-learn", "MLflow", "NumPy", "Pandas", "MLOps", "model serving", "NLP", "computer vision", "LangChain", "DSPy" ] }, { "name": "Observability", "keywords": [ "Prometheus", "Grafana", "Loki", "Promtail", "monitoring", "incident response" ] }, { "name": "Back-end & Databases", "keywords": [ "FastAPI", "Flask", "Django", "Django Rest Framework", "ClickHouse/Apache Doris", "PostgreSQL", "MySQL", "MongoDB" ] } ], "certificates": [], "volunteer": [], "custom_sections": [ { "title": "Projects", "entries": [ { "title": "pymolt", "subtitle": "Python Migration Tool", "start_date": "2026", "end_date": "", "url": "https://pymolt.zeelex.me", "highlights": [ "Runtime tracing tool built on sys.setprofile with streaming JSONL and a diff engine, detecting behavioral changes across dependency version upgrades.", "Architecture: pluggable TraceBackend, JsonlSink/MemorySink, three-level early-exit cascade for tracing overhead, TraceScope isolation." ] }, { "title": "MathMegaMind (M3), how machines can understand mathematics", "subtitle": "", "start_date": "2026", "end_date": "", "url": "", "highlights": [ "Testing whether systematic arithmetic depends on an agent's interface rather than model capacity.", "Built a graph-rewriting substrate (interaction combinators, Python, ~235k rewrites/s) with no numeric primitives; ruled out four stabilization mechanisms.", "Migrating to conservation-law dynamics in chemlambda, per convergent evidence from two independent substrates.", "Target: spontaneous emergence of persistent incrementable linear structures — an observer-independent criterion." ] }, { "title": "Brixel / companion & engine, local-first webassembly micro-app runtime", "subtitle": "", "start_date": "2026", "end_date": "", "url": "", "highlights": [ "Local-first execution engine translating natural language prompts into sandboxed micro-apps composed from verified WebAssembly modules, eliminating subscription bloat and data leaks.", "Zero-trust capability model blocking network, filesystem, and system APIs by default with deterministic CPU fuel metering and memory-capped threads.", "Architecture: Wasmtime host runner, .brix package spec, WASI syscall virtualization, zero-copy linear memory boundaries, and offline SQLite CRDT vault." ] } ] } ], "section_titles": [ [ "Profile", "Profile" ], [ "Work", "Work Experience" ], [ "Education", "Education" ], [ "Skills", "Skills" ], [ "Certificates", "Certifications" ], [ "Organizations", "Organizations" ] ], "section_order": [ "Profile", "Work", "Education", "Skills", "Certificates", "Organizations", { "Custom": 0 } ] }