Case Study 10 · Agriculture AI & Biosecurity
Poultry RAG
A veterinary decision-support and triage system with multimodal disease matching and automatic quarantine alerts
Intro
Product: Veterinary RAG and triage assistant, deployed as a Telegram bot.
Industry: AgTech / Veterinary Medicine
Target market: Smallholder poultry farmers, agricultural extension workers, and backyard keepers.
Role: Lead AI & Systems Developer responsible for multimodal retrieval pipeline, safety triggers, vet-review gate, and ingestion workflow.
What Made This a Good Bet
Smallholder poultry farmers lack rapid access to licensed veterinarians, leading to delayed responses to highly contagious disease outbreaks.
High-risk avian epidemics (like Avian Influenza and Newcastle Disease) spread rapidly, wiping out entire flocks and local farm economies within days.
Generic LLMs frequently hallucinate veterinary diagnoses, offering unsafe dosage numbers or useless home remedies.
Visual symptoms in sick birds are highly ambiguous, requiring structured reasoning over multi-modal vision inputs.
Veterinary support is time-critical; flock-level quarantine must be initiated instantly to prevent widespread transmission.
Certain avian diseases are zoonotic and notifiable, meaning they pose a severe public health risk and require immediate reporting to state authorities.
Ensuring diagnostic accuracy and treatment safety is paramount; unvetted medical steps can violate biosecurity regulations.
What I Built
Built a semantic search retriever in Qdrant to match natural language user descriptions against a vet-approved poultry disease knowledge base.
Integrated Gemini Multimodal Vision to analyze photos of sick birds and extract physical symptoms to supplement text query vector generation.
Engineered an automated safety filter that matches disease metadata and overrides the LLM output with emergency quarantine protocols if notifiable epidemics are detected.
Implemented a strict confidence-gating threshold that defaults to veterinarian escalation if the system matching score falls below 75%.
Created a robust vet-review validation framework in ingestion scripts to block unreviewed disease records from production.
The Stack
Beyond The Headline Metrics
Indexed 8 major infectious poultry diseases, including Avian Influenza and Newcastle Disease, with structured signs.
Achieved 100% recall on notifiable disease warnings on evaluation test suites.
Maintained rapid end-to-end response delivery under 2.0 seconds.
Established a mandatory verification gate, ensuring 100% of production data is reviewed by licensed veterinarians.
