All workFILE 11 / 16

Agricultural research

Grading nine thousand research records

Shipped at CGIAR · led by Nikhil Bery

01

The problem

Thousands of research records had to be assessed for real-world impact, a task too large to do by hand.

02

The constraint

Thousands of records had to be graded for real-world impact consistently, a volume too large to assess by hand.

03

The system

An impact-assessment pipeline over 9,166 research records with six fine-tuned classification models across five strategic impact areas, GPU-accelerated PDF layout detection, and Power BI dashboards.

FIG. 01 · SYSTEM FLOW · 4 NODES · 3 FLOWS

9,166 research records → GPU PDF layout detection. GPU PDF layout detection → 6 fine-tuned classifiers · 5 impact areas. 6 fine-tuned classifiers · 5 impact areas → Power BI dashboards

04

The outcome

9,166 records classified across five impact areas by six purpose-fit models, replacing manual assessment with a dashboarded pipeline.