GrafikInc. builds the infrastructure layer for reaching people the smartphone economy assumes don't exist — no app, no data plan, no smartphone, no literacy requirement. The delivery rail — USSD, voice with local-language text-to-speech, and a patent-pending platform for delivering AI over any SIM — is the reusable part. The reasoning layer on top, which fuses whatever signal exists into an actual answer instead of a generic lookup, is what's proprietary and hard to copy.
We started with one vertical — agricultural advisory for smallholder farmers in East Africa. It became something bigger: a reusable rail that unrelated verticals are now being built on top of. Agriculture is live. Others are in pilot, in concept, or in early conversation — and we label every one of them honestly, by the stage it's actually at.
The frontier AI now answering a CEO's question in New York can answer a farmer's on a $15 phone in Murang'a — but only if something carries it there. We build the layer that carries it: the intelligence rail for the people being left behind.
The share of low-income U.S. seniors without a smartphone is statistically identical to the share of rural Kenyans without one. The digital divide is not a developing-world problem. It is a poverty problem. One rail serves both.
Note: the 95% and $406B figures above describe the scale of the problem and a separate robot-companion pilot run by New York State — not GrafikInc. results. Our own senior work is an open-source project, not a live deployment (see Verticals).
We started with USSD, the protocol that runs *384# menus on any phone on any network. We built a bilingual advisory system for farmers in East Africa: no smartphone, no data, no app. It worked. We proved the thesis, hit the ceiling, and open-sourced the architecture.
The real breakthrough wasn't the advisory. It was realising the delivery layer was still borrowed. USSD depends on carrier gateways that fail, rate-limit, and charge per session.
So we built our own delivery layer that operates below the app layer, on 2G, on every network, in every country. The phone never talks to the AI directly — it talks to a shortcode.
No app. No data plan required.
Once the transport layer was solved, the question changed. It was no longer "how do we reach farmers?" It became: who else does this reach?
The same architecture that delivers a crop advisory in Dholuo to a farmer in Murang'a can carry a check-in to an isolated senior, a cold-chain handoff, or a labour-compliance log. Healthcare. Civic services. Financial guidance.
It is not an app. It is infrastructure. The verticals are what we build on top of it.
Each vertical below runs on the same delivery layer. Each is labelled by the stage it is actually at — live, in pilot, or concept. We don't round any of them up.
Agricultural advisory and regional pest & disease early-warning for smallholder farmers in Kenya. Delivered over 2G USSD, voice with local-language TTS, and — as of today — Telegram and WhatsApp, where the photo-diagnosis flow reasons against each farm's own conditions instead of trusting a raw classifier score.
View vertical →A routing and chain-of-custody layer for cold-storage logistics: fuel-optimised collector routing, USSD-witnessed dual-custody handoffs at the farm gate, and a harvest ledger that builds smallholders' first bankable credit history. In active pilot discussions with an East African cold-storage operator — not yet signed, not yet deployed.
View vertical →SMS-based AI companionship for isolated older adults without a smartphone — a direct architectural branch of the AgroFutures gateway, proving the rail generalises across unrelated populations and continents. Today it is an open-source reference implementation with no deployed users; we're seeking a US partner.
View vertical →A USSD labour-logging, wage-verification, and origin-traceability system for East African fibre cooperatives, aimed at EU CSDDD, CSRD, and ESPR / Digital Product Passport compliance. Architecture and a demo dashboard exist; no live deployment or signed cooperative partner yet.
View vertical →Because the rail is domain-agnostic, interest is arriving from places we didn't build for. We're in early conversations with a national government, a national agriculture ministry, a European defence ministry, and women's-health and mental-health NGOs in East Africa. These are inbound interest, not pilots or deployments — nothing is built or signed for them yet. We mention them only as evidence that one rail reaches across unrelated domains.
The same 2G connection that delivers a farm advisory can also capture registrations, pickup logs, labour hours, and abuse reports. Every upstream log can feed back into the intelligence layer, so the pipe improves as it runs.
What follows is what the architecture enables, not a description of deployed systems. The rail is live for agriculture. The cold-chain application is at pilot stage; the labour-compliance application is at concept stage. Each capability below is tagged accordingly.
The Rail
A delivery layer that runs on 2G, on any carrier network, in any country: USSD and voice with local-language text-to-speech, plus a patent-pending platform for delivering AI knowledge over any SIM — no app, no data plan, no smartphone. It works where every other AI interface fails.
The reasoning layer runs any capable model on the backend. The rail is the reusable, hard-to-replicate part.
The Reasoning Layer
The layer on top doesn't return a raw score. It fuses whatever signal exists — a text report, a photo, a farm's conditions, a patient's context — into an actual answer. Its discipline is confirm, discard, or flag:
never silently trust a classifier score that doesn't fit what it already knows, and never silently reject an out-of-pattern signal either. When the evidence conflicts, it hedges and asks for more rather than forcing a false answer in either direction.
2G
Any carrier network
~1–2KB
Total session data per advisory
3B+
Addressable feature-phone users
The USSD and SMS architecture is live. Our next-generation delivery layer, which removes the last carrier dependency, is in active development. Partnership and research inquiries welcome.
The delivery infrastructure is MIT licensed. We believe the protocols that connect excluded populations to AI should be owned by no one. The intelligence models, biological simulations, and advisory engines built on top of them are proprietary.
Fork these repos to build your own applications on 2G infrastructure. Build for your market, your language, your population.
We work with researchers, NGOs, development agencies, and funders building toward AI access for excluded populations.
For partnerships, grant co-applications, or technical collaboration:
grafikinc.com
Sent.
We'll be in touch.
Something went wrong. Email us directly: [email protected]