About the role
Position overview:
Build the “Open Data Spine” that powers Symbioverse’s Regenerative AI Engine. You will design a reproducible ingestion + harmonisation pipeline for earth observation, climate, and soil baselines, and deliver a versioned, ML-ready feature store aligned to a canonical Field–Season schema.
Problem statement (why this exists):
Regenerative decision-support fails without robust spatiotemporal inputs. Open EO/climate/soil data exists, but it is fragmented, inconsistent in units/projections, and hard to reproduce. This track turns messy open data into a reliable, versioned substrate that downstream modelling, causal, and planning modules can trust.
Data you will work with:
- Primarily open/public sources (e.g., Sentinel-1/2, ERA5 and/or NASA POWER, SoilGrids, terrain/landcover layers as needed).
- Optional internal data: Symbioverse field datasets (Spain) may be available under NDA for validation and stress-testing (not required).
Key responsibilities (linked to the innovation project):
- Implement ingestion for farm/vineyard-relevant open sources (EO, climate, soil baselines).
- Define data contracts: fields, units, provenance, missingness rules, and versioning strategy.
- Produce an ML-ready feature store (e.g., Parquet/DuckDB/Arrow) and clear documentation.
- Deliver a benchmark modelling task + evaluation harness (baseline model, metrics, error analysis) to unblock downstream modules.
Why your expertise is crucial:
This role requires strong geospatial/time-series engineering plus evaluation hygiene—without it, every downstream model will leak, drift, or break.
Key outcomes for the Visiting Talent:
- A portfolio-grade, end-to-end data engineering deliverable (pipeline → feature store → benchmark).
- Demonstrated ability to ship reproducible artefacts (tests, docs, containerised run).
- Experience bridging open geospatial data into applied AgriTech/ClimateTech modelling.
Stream:
- Deep-Tech Talents (research-focused).