Daniel H.

I take ownership of production ML models, LMM, RAG, algorithms and the pipelines around them. I reverse engineer inherited systems, document assumptions and constraints, make runs reproducible, then add pragmatic MLOps: model registry, monitoring, data validation, and safe rollout. My background is environmental science plus engineering, so I’m comfortable with remote sensing style imagery, large rasters, and noisy real world inputs. Inheriting and improving existing ML systems - Reproduce training and inference e2e (from raw imagery to outputs) - Document model behaviour, failure modes, constraints, and performance - Build a lightweight model registry (model cards + dataset and config provenance) - Add monitoring: service health, data integrity, drift / out of domain detection - Define retraining triggers, evaluation gates, and safe promotion / rollback My strength is end to end delivery. I can start from raw sources, remote sensing products, field or sensor datasets, or AIS style tracking data, then build the full workflow: cleaning and QA, spatial processing and feature engineering, modelling (including computer vision), and finally a dashboard or API ready output that stakeholders can use. I write clear documentation and explain assumptions so your team can run and maintain the work after handoff. What Ive delivered - AI/LLM Engineering - Multi-agent orchestration platform Haiku for fast tasks), intelligent routing, SQLite backed memory with cross-agent recall/handoff, priority command queues, and a heartbeat daemon for autonomous scheduled operations. Daily work with the Anthropic SDK, prompt engineering, and MCP server development. - AI Powered Automation - LLM scored relevance pipelines (batched Haiku calls), automated research systems (data fetch → LLM analysis → actionable output), Telegram bot with agent routing and direct-API fallback, - voice pipeline (Whisper STT → agent → Edge-TTS). These systems run without human input. Geospatial pipelines and analysis: PostGIS, QGIS, xarray, rasterio - Machine learning and predictive modelling: Python, PyTorch, TensorFlow, scikit learn, MLflow, spatiotemporal features, evaluation - ML architecture, computer vision, and complex data correlation. - Dashboards and reporting: Streamlit, Tableau, stakeholder ready visuals and summaries - Cloud and automation: AWS or GCP, Docker, Airflow, n8n, reproducible runs and deployment friendly code - If your project involves making AI do useful work reliably from model training to LLM integration to deployed pipelines I can take it from concept to production. - CI/CD How I work: I scope clearly, communicate early if something shifts, and deliver working systems with documentation. I'm comfortable across the stack - model training, API integration, infrastructure, and the glue code that makes AI systems reliable in production. If you need someone who can translate environmental complexity into a working system with clean outputs, I can help you take the project from concept to delivery with a focus on accuracy, efficiency, and maintainability.