Evidence‑Grounded Agronomy Copilot & MRV Metrics
Description
Position overview:
Deliver the user-facing and integration layer: a grounded agronomy copilot that produces evidence packs (citations + retrieved snippets + quantitative outputs) and an MRV-style proxy dashboard from open data. This track also owns integration: wiring Tracks A–E into one runnable prototype with an integration test harness.
Problem statement (why this exists):
If an AI system cannot be audited, it will not be trusted in agriculture. This track builds a “show-your-work” assistant that connects recommendations to evidence and measurable proxies—while enforcing evaluation and guardrails against hallucination.
Data you will work with:
- Outputs from Tracks A–E (feature store, embeddings, models, causal effects, KG).
- Curated public technical sources for retrieval (papers, guidelines, standards) with strict citation requirements.
- Open data streams for MRV-style proxies; optional internal field data for validation (not required).
Key responsibilities (linked to the innovation project):
- Build a retrieval-augmented generation (RAG) workflow grounded in the KG (Track E) + curated technical sources.
- Implement guardrails and evaluation: groundedness, citation accuracy, hallucination testing, user-task metrics.
- Design MRV proxy metrics computable from open data (yield stability proxy, soil carbon proxy trend, water/input reduction proxy) with uncertainty.
- Create an integration harness that runs the end-to-end demo (containerised) and verifies module interfaces.
Why your expertise is crucial:
This role requires systems thinking: interface discipline, evaluation, and integration mindset—without it, the engine remains a set of disconnected notebooks.
Key outcomes for the Visiting Talent:
- An auditable, evidence-grounded assistant prototype + evaluation suite.
- Hands-on experience with integration and quality gates for AI systems.
- A demonstrable prototype suitable for deep-tech product narratives.
Stream:
- Deep-Tech Talents (research-focused).
Skills
Sign in to read and join the discussion.