The Smallholder Stack: Why a Nigerian Poultry + Catfish Farm Was the Right First Customer for an AI Operations SaaS

A developer built FarmOps Desk, an AI-powered operations SaaS for smallholder livestock farms, and chose a Nigerian poultry and catfish farm as its first customer. Nigeria's dense poultry and catfish sectors, robust payment infrastructure like Nomba, and English-language operations made it a strategic launch market. The architecture generalizes across smallholder economies in Sub-Saharan Africa, Southeast Asia, South Asia, and Latin America, with design decisions like storing monetary values as smallest integer units and voice-first interfaces.

Part of the H0: Hack the Zero Stack submission. See the project on Devpost. Nigeria is the launch market. It is not the ceiling. This article explains the reasoning behind that distinction, and why the architecture I built for FarmOps Desk generalizes across the smallholder livestock economies of Sub-Saharan Africa, Southeast Asia, South Asia, and Latin America. The product decisions Nigeria drove are durable; the marketing decisions are not. The UN Food and Agriculture Organization estimates that smallholder farmers produce up to 80% of the food supply in developing countries and depend on livestock for a meaningful share of their income. The World Bank's IFC puts the smallholder agritech market at $510 billion annual spend across the regions where smallholders are the dominant producer. The operational problems are uniform across these regions: These are not "Nigerian problems." They are smallholder livestock problems , expressed in different currencies and languages across a $1.4 trillion addressable market IFAD 2024 estimate, livestock sub-sector . Three reasons, none of which are sentimental: 1. Density. Nigeria is the 4th-largest poultry producer in the world by headcount after China, Brazil, USA . It's also the fastest-growing catfish producer in Sub-Saharan Africa. The addressable farm count is in the millions. 2. Proven Payment Infrastructure. To build a B2B SaaS, you need reliable ways to process payments locally. Nigeria has robust payment providers like Nomba, which offer modern APIs like webhooks and idempotency similar to what Stripe uses globally. Because I built the payment layer to support Nomba's standard webhook pattern, integrating future local payment rails like M-Pesa in Kenya or Pix in Brazil will just require dropping in a new provider file, not rewriting the billing engine. 3. English-language reduces go-to-market friction. The AI's grounding prompts, the UI copy, and the demo video are all in English. For a solo dev or small team, this is significant. I can ship a Kenyan or Ghanaian version next month without re-recording or re-translating. For non-English markets like Indonesia and Brazil, the farms.language column already supports per-farm localization; the system prompt adapts. Nigeria is a market of 200M people with a real poultry + catfish sector, real payment infrastructure, and English-language operation. Picking it first is a business decision, not a sentimental one. Every Nigeria-driven decision turned out to be a global-smallholder decision in disguise. This is the architecture's strongest claim to scalability. NGN has 100 kobo per naira. I store all monetary values as BIGINT kobo. The CHECK constraints amount minor = 0 , balance after = 0 work in any currency — IDR rupiah no minor unit, just store rupiah as the minor , BDT paisa, PKR paisa, BRL centavos. The pattern is "smallest integer unit" — a Nigerian decision that generalizes to every currency on the planet. The alternative — NUMERIC 10,2 float money — is the wrong answer everywhere, but most pronounced in markets with high inflation where small units matter. Nigerian smallholder farmers are not all low-literacy, but the operational reality is: you're standing in a broiler house at 5am with two hands full of feed, and you don't have time to type. Voice is the natural interface. The same is true in rural Indonesia, Bangladesh, Kenya, Brazil. The use-voice-session.ts component, the Bedrock Nova Sonic bidirectional session, the audio fallback via MediaRecorder — all of this generalizes. The system prompt language adapts from farms.language . A 200-bird subsistence farm in Kaduna has different risk tolerance than a 50,000-bird commercial operation in Lagos. The autonomy tier suggest | draft | auto lets each farm dial in. The same dial serves a 50-bird farm in Java and a 5,000-fish pond in Vietnam. The crucial sub-decision — financial and destructive writes hardcoded to draft regardless of tier — is universal. No farmer anywhere wants the AI to record a sale or a mortality event without confirmation. The autonomy floor is a safety invariant, not a cultural preference. lib/payments.ts has a provider interface with a mock implementation and a nomba implementation. Adding Stripe for international cards , M-Pesa Kenya , Pix Brazil , or GoPay Indonesia is one provider file each. The webhook + idempotency pattern is identical across all of them; the diff is the request/response shape. lib/messaging/provider.ts abstracts Termii Nigeria SMS vs Resend email . For Kenya, Africa's Talking slots in. For Indonesia, WhatsApp Business API. For Brazil, Twilio. The interface is the same. The schema has farms.currency , farms.language , farms.timezone . Today these are mostly set to NGN/en/Africa/Lagos in production. But the assistant system prompt is built per-request from farms.language . The daily insight schedule uses farms.timezone . The money formatter reads farms.currency . A farm in Bangladesh or Brazil works today, with no code change. production units.species is a TEXT column, not an enum. Today the UI surfaces broiler and catfish . The schema accepts goat , cattle , pig , duck , tilapia without modification — only the species catalog and the breed lists need updating. Nigeria's narrow focus is a UI choice, not a schema constraint. Because the core architecture is decoupled from regional specifics, expanding to new markets becomes a data-entry exercise rather than a software rewrite. Here is how that scales: | Year | Market | TAM | What changes from the previous market | |---|---|---|---| | 2026 | Nigeria | $8B poultry + catfish | Launch market. Nomba payment, English UI. | | 2027 | Kenya, Ghana | $5B poultry | Add M-Pesa payment rail. Same language. Same species. | | 2027 | Indonesia, Philippines | $25B poultry + fish | Add GoPay / GCash payment. Bahasa Indonesia / Tagalog language pack. Same species. | | 2028 | Brazil, India | $40B poultry + fish | Add Pix / UPI payment. Portuguese / Hindi language pack. Same species. | | 2029+ | Vietnam, Bangladesh, Egypt | $30B+ | Same pattern. | The key takeaway here is that entering a new market requires just four predictable updates: the currency, the payment provider, the language pack, and the local species/breed list. None of these require database schema changes. The cost of entering a new market is simply data onboarding and a single payment integration, not a platform rewrite. The reason this matters more than the architecture is what the architecture is for. Nigeria's poultry sector loses an estimated 15-20% of broiler production to preventable causes — heat stress, vaccination gaps, feed mismanagement. That's roughly ₦600 billion $750M per year in dead birds. A 5% reduction across the sector would be $375M/yr in farm income preserved . The numbers scale across the regions: a 5% mortality reduction across Indonesia's poultry sector is another $500M/yr. Across Brazil's, $400M/yr. The aggregate smallholder livestock TAM for AI-assisted operations is in the tens of billions per year in recovered production, before counting the productivity gains. This is the real-world impact that justifies the engineering effort. The architecture I described in the companion pieces — Aurora + pgvector + Bedrock, keyless OIDC, atomic credit reservation — is just the substrate. The reason to build it is the smallholder farmer who stops losing birds to a 4am heat wave because the assistant drafted a ventilation task the night before and the farmer confirmed it with one tap. Core Features Implemented: Future Roadmap: The schema doesn't know it's about Nigeria. The currency column says NGN today and would say IDR or BRL or BDT tomorrow. The species column accepts any string. The payment interface accepts any provider. The language column drives the assistant's system prompt. The autonomy tier respects every culture's risk tolerance by giving each farm the dial. The product launched in Nigeria because that's where the densest cluster of smallholder poultry + catfish farmers with real payment rails and English fluency happens to be. The architecture launched for the world. Built for the H0 Hack the Zero Stack hackathon H0Hackathon , Monetizable B2B App track. Deployed on Vercel, Amazon Aurora PostgreSQL as primary backend, Amazon Bedrock for AI. Nigeria is the launch market, not the ceiling. I created this piece of content for the purposes of entering the hackathon.