← Back to CV

AI · 2024-12 → present

CGIAR Agricultural Risk Intelligence Tool

Associate Team Leader — Digital Platforms · Alliance of Bioversity International & CIAT (CGIAR)

AI risk-scoring view: subcategory score bars (revenue, cost, debt, cash flow, controls).

The problem

CGIAR teams assessed the risk of agribusinesses by hand — reading scattered financial, market and climate data and writing a report per business. It didn't scale, and two analysts could reach different conclusions from the same data.

The approach

A risk-intelligence tool that ingests an agribusiness's data and produces a documented, evidence-backed risk report. Several models on AWS Bedrock extract evidence, reason over it, and score the business across subcategories — revenue stability, cost structure, debt levels, cash flow, financial controls — each rated and explained, with concrete recommendations. The human reviews and approves; the machine drafts.

Architecture

  • Serverless on AWS: Lambda + API Gateway, Bedrock for the models, S3 / DynamoDB for storage.
  • A RAG pipeline over financial, market and climate sources feeds the reasoning.
  • Typed end to end (NestJS + Next.js + TypeScript, with Python for data work).

The outcome

  • Per-business analysis drops from hours of manual work to minutes of automated scoring.
  • Consistent, comparable reports instead of free-form write-ups.
  • The pipeline pattern was reused across other internal CGIAR tooling.

Public documentation: the risk-analysis tool docs.

Stack

  • AWS Bedrock
  • Anthropic Claude
  • RAG
  • NestJS
  • Next.js
  • TypeScript
  • Python
  • Lambda
  • API Gateway
  • S3
  • DynamoDB

Outcome

  • Hand analysis per business collapses into a few minutes of automated scoring.
  • Consistent, evidence-backed risk reports instead of free-form write-ups.
  • Reusable AI pipeline pattern reused across other internal CGIAR tooling.