LivestockAI

Inspiration

I began LivestockAI because I have lived on both sides of the problem. I am a software engineer, but I am also a Nigerian broiler and catfish farmer. On a farm, a problem rarely announces itself as a neat software notification. It appears as birds eating slightly less, water measurements drifting, feed finishing sooner than expected, mortality changing, or several small records that do not look important until they are considered together. Many farmers manage those signals through memory, paper, and WhatsApp while professional support is far away.

The African Union estimates that about 33 million smallholder farms produce roughly 70% of Africa’s food. These farmers are not short of determination. They are short of timely attention, organized evidence, and affordable access to specialist knowledge. I wanted to build the AI farm manager I needed: one that understands the farm’s own records, watches for what deserves attention, and helps the farmer take the next step without taking control away from them.

What it does

That became LivestockAI and its AI farm manager, Farmhand. Farmers record mortality, feed, weight, water quality, health, expenses, and sales. Farmhand then connects those records across six Gemini-powered surfaces: grounded chat, real-time voice, daily planning, image-assisted triage, what-if analysis, and monitoring.

Sentinel is the clearest example of what makes the system AI-native. A six-hour scheduler evaluates configured rules and recent farm data even when the farmer has not opened a chat. It persists findings and their severity. Gemini can then turn the underlying evidence into a clear explanation and a suggested next step. The system can notice and prepare work without waiting to be prompted.

LivestockAI also connects the people around the farm. Extension officers manage bounded portfolios of approximately 20–30 assigned farms, while supervisors triage risk across several officer portfolios. Buyers can use grounded AI to understand listings and contracts. Veterinarians can review farm and health context before consultations. Suppliers can manage catalogues and orders, and transporters can manage jobs and delivery evidence. The Credit Passport turns consented operational history into a structured dossier that can support future lender evaluation.

How we built it

I built the current application from its first repository commit on June 13. The backend is FastAPI and PostgreSQL; the web application is React and TanStack; the deployed stack runs on Google Cloud Run with Cloud SQL, Cloud Build, Artifact Registry, Secret Manager, and Google Cloud observability.

Google Gemini powers the application through the official Google Gen AI SDK. Twenty-eight domain read-tool modules ground the AI in live operational data. FarmScope and DistrictScope enforce server-side isolation, while typed tool inputs, timeouts, metering, audit logs, and approval requirements constrain execution.

AI also helped me build the company. I used Codex, Antigravity, Agy CLI, Orca, ZCode, and Muse Code for implementation, refactoring, testing, debugging, documentation, and review. I remained responsible for architecture, product decisions, evidence integrity, agricultural safety, and accepting changes.

Challenges we ran into

The hardest problem was balancing autonomy with accountability. Agriculture is too consequential for uncontrolled automation: an incorrect medication action can harm a flock, and an incorrect financial action can harm a family business. Farmhand can therefore identify a need and prepare a typed action draft, but medication, money, marketplace commitments, and destructive changes remain subject to human approval.

The second challenge was serving rural users without designing only for ideal connectivity or expert software users. LivestockAI is installable and optimized for lower-bandwidth use, with five interface locales and voice interaction. The third challenge was cost: model routing, metered credits, context reuse, and surface-specific limits allow me to understand and control AI cost as usage grows.

Accomplishments that we're proud of

In roughly two months, I built and deployed a broad agricultural operating system with 42 database migrations, six AI surfaces, 28 governed read-tool modules, scheduled Sentinel monitoring, farmer and extension workspaces, marketplace roles for buyers, vets, suppliers, and transporters, Credit Passport, billing, communications, and observability.

I am especially proud of the quality loop. A separate Gemini judge scores sampled responses for specificity, actionability, safety, localization, and coherence. A weak result can produce a pending learning proposal, but that proposal is never automatically applied. People review and approve changes, so the system can improve without silently rewriting its own safeguards.

What we learned

I learned that the strongest agricultural AI is not the one that claims the most autonomy. It is the one that notices important signals early, grounds its reasoning in the farm’s real data, clearly communicates uncertainty, and knows when a farmer, veterinarian, extension officer, or administrator must decide.

I also learned that the farmer subscription is only one route to impact. Government programs, LGAs, cooperatives, and NGOs can sponsor access while extension teams use the same platform to direct limited field capacity. That model can make advanced software available to smallholders without forcing the smallest farmer to carry the full cost.

What's next for LivestockAI

The platform is deployed, and I am beginning farmer onboarding today. I will measure activation, record completeness, retention, response to Sentinel findings, cost per active farm, and conversion to the Grower plan. Within the next two months, I plan to secure a bounded LGA, state agricultural program, cooperative, or extension pilot with clearly assigned farms and measurable outcomes.

From there, I will validate Credit Passport with financial partners and the buyer, veterinary, supplier, and transporter workflows with real marketplace participants. My goal is to help small farms become stronger businesses while making scarce human expertise reach them earlier. The potential is continental, but it begins with one deployed system, one farmer cohort, and evidence that compounds with every production cycle.