Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Building AI-based advisory services for smallholder farmers: Technical learnings from the AIEP Initiative

Stewart Collis, Florence Kinyua, Vikram Kumar, Howard Lakougna, Christian Merz, Kirti Pandey, Christian Resch

TL;DR

The paper analyzes technical learnings from five AI-based advisory MVPs for smallholder farmers in Kenya and Bihar, reporting strong early user satisfaction (NPS ~60) and a modular architecture that couples voice-enabled interfaces with a reasoning layer that fuses LLMs, external data, and RAG over curated agricultural content. It highlights persistent challenges in latency, low-resource language support, and labor-intensive content curation, while proposing enablers such as data sharing, common corpora, improved language AI, and robust evaluation benchmarks. Practical guidance is offered for building such systems, including multidisciplinary teams, content governance, multi-channel strategies, reusing existing technologies, and early feedback loops, along with the introduction of golden Q&A datasets for agricultural LLM evaluation. The paper also discusses the future trajectory of AI-based advisory, balancing advances in end-to-end LLMs with the need for auditability and scalable digital public goods infrastructure to ensure safe, impactful deployment at scale.

Abstract

We report technical learnings from five AI-based agricultural advisory MVPs deployed in Kenya and Bihar, India, under the AIEP Initiative. A 800-farmer study found high user satisfaction (NPS ~60). All solutions implement a modular two-part architecture: (i) an interface component (IVR /WhatsApp / app) with ASR-MT-TTS for multilingual voice access; and (ii) a reasoning component combining LLMs capabilities with query orchestration, external data (weather/soil/markets), and RAG over curated agricultural corpora. We describe key challenges: (a) latency, especially for voice; reductions were achieved via in-country hosting and audio minimization, but consistent <5s remains challenging; (b) language coverage: low-resource ASR/MT integration and nonstandard scripts hinder end-to-end quality; and (c) corpus curation: access, validation, and maintenance are labor-intensive, as well as provide recommendations on how to develop similar systems. We discuss common enablers including (a) data sharing, (b) common corpora, (c) better language AI and (d) evaluation and benchmarking. We also present golden Q&A sets to evaluate LLM capabilities for smallholder agriculture.

Building AI-based advisory services for smallholder farmers: Technical learnings from the AIEP Initiative

TL;DR

The paper analyzes technical learnings from five AI-based advisory MVPs for smallholder farmers in Kenya and Bihar, reporting strong early user satisfaction (NPS ~60) and a modular architecture that couples voice-enabled interfaces with a reasoning layer that fuses LLMs, external data, and RAG over curated agricultural content. It highlights persistent challenges in latency, low-resource language support, and labor-intensive content curation, while proposing enablers such as data sharing, common corpora, improved language AI, and robust evaluation benchmarks. Practical guidance is offered for building such systems, including multidisciplinary teams, content governance, multi-channel strategies, reusing existing technologies, and early feedback loops, along with the introduction of golden Q&A datasets for agricultural LLM evaluation. The paper also discusses the future trajectory of AI-based advisory, balancing advances in end-to-end LLMs with the need for auditability and scalable digital public goods infrastructure to ensure safe, impactful deployment at scale.

Abstract

We report technical learnings from five AI-based agricultural advisory MVPs deployed in Kenya and Bihar, India, under the AIEP Initiative. A 800-farmer study found high user satisfaction (NPS ~60). All solutions implement a modular two-part architecture: (i) an interface component (IVR /WhatsApp / app) with ASR-MT-TTS for multilingual voice access; and (ii) a reasoning component combining LLMs capabilities with query orchestration, external data (weather/soil/markets), and RAG over curated agricultural corpora. We describe key challenges: (a) latency, especially for voice; reductions were achieved via in-country hosting and audio minimization, but consistent <5s remains challenging; (b) language coverage: low-resource ASR/MT integration and nonstandard scripts hinder end-to-end quality; and (c) corpus curation: access, validation, and maintenance are labor-intensive, as well as provide recommendations on how to develop similar systems. We discuss common enablers including (a) data sharing, (b) common corpora, (c) better language AI and (d) evaluation and benchmarking. We also present golden Q&A sets to evaluate LLM capabilities for smallholder agriculture.
Paper Structure (27 sections, 8 figures)

This paper contains 27 sections, 8 figures.

Table of Contents

  1. Introduction: What is this about?
  2. Background on the AIEP initiative
  3. Solutions descriptions: What are the 5 MVPs?
  4. Related work: What other solutions are there?
  5. General architecture: How do AIEP advisory systems work?
  6. A walk through
  7. The interface
  8. The reasoning component
  9. Learnings
  10. Interface
  11. Latency
  12. Reasoning Component
  13. How to approach building AI-based advisory?
  14. The Team
  15. Monitoring and measuring
  16. ...and 12 more sections

Figures (8)

  • Figure 1: Common high-level architecture of AIEP MVPs
  • Figure 2: Stylized theory of change and AIEP Standard Indicators
  • Figure 3: ASR Performance in Gikuyu as evaluated by Gooey.ai
  • Figure 4: Combined Performance of ASR and MT in Gikuyu as evaluated by Gooey.ai
  • Figure 5: LLM performance on agricultural questions as rated by human agronomists
  • ...and 3 more figures