README.md

metadata

title: KALRO Maize Research Assistant
emoji: 🌽
colorFrom: green
colorTo: green
sdk: docker
pinned: false
license: cc-by-4.0

KALRO Maize Research Assistant

A research chatbot for Kenya Agricultural and Livestock Research Organization (KALRO), grounded in 16 research papers and training manuals on Kenya maize production.

Knowledge Base

• Hybrid breeding and combining ability (KALRO Muguga inbred lines)
• Genotype × environment interaction and stability analysis
• Water harvesting technologies for ASAL (zai pits, ngolo pits, tied ridges)
• Maize Lethal Necrosis (MLN): epidemiology, resistance screening, seed transmission
• Genetic gains in the Kenya Hybrid Maize Program (1999–2020)
• Farmer preferences for stress-tolerant variety traits
• Agronomic factors and the smallholder yield gap
• KCEP-CRAL and KCSAP extension programmes

How It Works

Questions are answered using a two-layer retrieval system:

1. Wiki layer — structured summaries of each paper (primary source)
2. PDF layer — raw extracted text from all 16 documents (fallback for detail)

Retrieval uses all-MiniLM-L6-v2 embeddings stored in a local Chroma vector database. Generation uses Claude Sonnet via the Anthropic API.

Setup (self-hosting)

Set ANTHROPIC_API_KEY as a Space secret (Settings → Variables and secrets).

To rebuild the knowledge base from new documents:

python3 chatbot/pipeline/run_pipeline.py