White Paper — Déblo

Real-Time Voice & Eyes AI

A platform thesis: created in Abidjan, deployed on Google Cloud Vertex AI, built for the world

Author: Juste Azandegbe Gnimavo, Founder & CEO, ZeroSuite Inc. — justegnimavo.com Version: 3.1 — May 2026 Classification: Public — Free for distribution (CC BY 4.0)

Executive Summary

We don't sell AI. We sell access to expertise — to 1 billion people who never had any.

Déblo is a real-time voice & eyes AI platform deployed on Google Cloud Vertex AI. The user speaks; Déblo answers live, like a phone call. The user shows their world through the camera; Déblo sees what they see, in real time, and the conversation continues. The platform is built for people who prefer speaking and showing over typing — a preference shared by parents, students, traders, freelancers, support teams, and communities across local-language contexts everywhere. Mobile money from 100 FCFA. First-screen camera. Voice as primary interface. Designed for the user who never had a Visa card, a keyboard, or an expert one phone call away.

The product was created in Abidjan, Côte d'Ivoire, through a U.S. company, and is built for global use. Africa is the launch beachhead, the cultural anchor, and the strongest initial adoption market — but the product itself is globally usable from day one. The same Vertex AI foundation — Gemini Flash Live for the real-time voice runtime, Gemini frontier models for multimodal reasoning and live camera understanding, open-weight Gemma in the memory and RAG layer, and Gemini Embedding 2 (released by Google on May 20, 2026) powering semantic search across user memories, tasks, and the RAG corpus — serves four markets:

1. Daily assistance — parents, students, traders, freelancers, families.
2. K-12 education — homework, explanations, revision, exams, OCR, voice tutoring, real-time camera reading of handwritten exercises.
3. Customer support — AI voice agents for support teams, call centers, and live assistance.
4. Languages and inclusion — translation, multilingual coverage, voice access for people who prefer speaking, real-time vision for mothers reading school reports, elders deciphering prescriptions, and traders verifying invoices.

Most consumer AI products today are text-first, English-dominant, and require an international bank card. Voice modes, when available, sit behind a paywall priced for high-income markets. Real-time camera vision — the AI sees what the user sees live — is essentially absent outside frontier consumer demos. Code-switching between French and a local African register is handled mechanically rather than conversationally. Mobile money is invisible.

Déblo addresses each of these gaps as a product foundation, not a localization layer. The voice runtime is real-time and runs as a continuous loop on Gemini Flash Live, with the camera streaming alongside. The language coverage matches the languages Africans actually use across borders for business, school, and government today — French, English, Spanish, Portuguese, Arabic, German, and Chinese — with a behavioural register layer that stays with the user when they mix in local expressions. The payment rail is mobile money from 100 FCFA per top-up; no bank card required. The cultural register handling was built in Abidjan, by an African team, against real African user input.

Wider native voice output in local African languages is a multi-year roadmap, contingent on grant-funded corpus building, native linguistic review, and regional voice synthesis fine-tuning. It is not a launch claim. It is the institutional partnership opportunity.

This white paper is the case for a real-time voice & eyes AI platform deployable across four markets on a single Vertex AI foundation — with Africa as the launch beachhead and global users from day one.

Key figures:

• 850 million voice-first addressable users across Africa and diaspora — the strongest initial market for the platform
• 7 global languages spoken fluently by Déblo at launch (French, English, Spanish, Portuguese, Arabic, German, Chinese)
• 38% average secondary school dropout rate in our six launch countries (Market 2)
• Unsolicited inbound from a regional ISP service manager on level-1 call center automation (Market 3)
• 10+ local African languages targeted for native voice output by 2028, contingent on grant funding (Market 4 institutional thesis)
• ~$0.15 USD entry-level Déblo session cost
• Day-one Google Cloud Vertex AI deployment
• Public launch June 1, 2026. Real traction numbers published 90 days post-launch (September 2026)

Section 1 — The real-time voice & eyes AI thesis

1.1 Where voice & eyes AI sits in 2026

Real-time voice AI reached production-grade quality in late 2025. Real-time camera vision — the model seeing the user's environment as a continuous stream rather than a screenshot — reached production-grade quality in early 2026, primarily through Gemini Flash Live on Vertex AI. Phone-call-like latency, paired with conversational camera understanding, is now feasible on entry-level hardware. The technical envelope that mainstream AI assistants reserve for paid voice + vision modes in English is, today, deployable in French, in a broader set of languages, on entry-level Android handsets, over mobile networks in low-bandwidth conditions, and on mobile-money payment rails.

The category remains largely structured around English-dominant assistants billed by international card. Real-time voice & eyes products with local cultural register handling, multilingual coverage matching cross-border African usage, and mobile-money distribution remain rare. That is the gap Déblo addresses — not as a regional copy of a global product, but as a platform built around the way voice-first users actually behave.

We surveyed AI-funded startups in May 2026. Africa appears rarely in product taglines. Real-time voice is mostly sold as developer tooling (voice infrastructure providers, voice runtime APIs, voice model fine-tuning services), not as a finished consumer product. Real-time camera vision is largely confined to frontier consumer demos by hyperscalers. The consumer-facing real-time voice & eyes product, built with mobile-money distribution and cultural register handling for code-switching users, is a category Déblo is choosing to occupy.

1.2 What "real-time voice & eyes" means, exactly

The word "real-time" is overused. Here it means a specific technical commitment:

• Low-latency, phone-call-like response time from the moment the user finishes a sentence to the moment Déblo's first audio token plays on device. Measured in production, on entry-level Android handsets, over Orange CI 4G during peak hours in Abidjan.
• No turn-taking awkwardness. The user does not press a button to speak or wait for a "your turn" beep. Déblo interrupts when appropriate, stays silent when it is not. The conversation flows.
• Native voice understanding, not speech-to-text then text-to-speech then back. The voice runtime handles speech and reasoning in one continuous loop on Gemini Flash Live.
• Live camera streaming, not screenshot upload. The user points the camera at an exercise, a prescription, a school report, an invoice; Déblo sees the stream and answers about what is visible in the same voice turn.

This envelope has matured into production on global AI mobile apps, where it sits behind premium subscription tiers payable by international card. Déblo delivers the same envelope in French, across the languages Africans use today, on entry-level hardware, on mobile-money rails, in an app designed for the way voice-first users actually behave.

1.3 The horizontal voice & eyes AI platform thesis

Déblo is built as a horizontal layer, not a single vertical app. The same real-time voice plus eyes plus reasoning stack serves:

• A primary-school student in Bouaké asking for help with a fractions exercise — held up to the camera.
• An accountant in Dakar drafting a SYSCOHADA-compliant balance sheet narrative.
• A market vendor in Cotonou calculating change at the till.
• A telecom subscriber calling Déblo as a level-1 support agent for a top-up issue.
• A government-deployed assistant explaining a tax form in a local language.
• A diaspora user in Brooklyn calling Déblo to translate a French letter held in front of the camera into English for an English-speaking spouse.
• A freelance designer in Casablanca dictating a client proposal between meetings.
• A mother in Lomé pointing her phone at her child's bulletin scolaire to hear it read aloud and explained.

These are not eight different products. They are eight configurations of one product. The voice & eyes platform is the same in every case; the markets differ in system prompts, knowledge bases, and pricing models.

4 markets. One foundation. This is the platform thesis.

1.4 Position in the African AI ecosystem

An African AI ecosystem is forming around language modelling for African languages, voice agents for specialised verticals, and regional NLP tooling. Déblo operates on an adjacent layer: a horizontal real-time voice & eyes consumer AI platform on Google Cloud Vertex AI, distributed via mobile money across the 6 launch countries, with 4 markets on a single foundation. The work across the broader ecosystem is complementary, and partnerships with African AI teams on language-specific fine-tuning and corpus collaboration are explicitly on the multi-year roadmap.

1.5 The accessibility moat — built for users who never read, sign up, or wait

The technical envelope is only half the story. The other half is the access point.

Mainstream consumer AI products require an email account, an OTP, a tutorial, a tier selection, and a payment method before the user reaches the first useful interaction. This funnel is designed for a literate, English-speaking, card-holding user. For the 1 billion adults the platform targets — many of whom cannot read, have never signed up for any online service, and would not understand a five-step onboarding slideshow — that funnel is the product.

Déblo collapses the funnel to zero. The first screen is the call screen. A welcome credit is already loaded. The two icons on that screen — the mic and the camera — are the only ones every human recognizes regardless of literacy, language, or device experience. There is no OTP, no slideshow, no tutorial, no tier selection. The user taps and speaks. Or taps and shows.

This is the second architectural moat alongside mobile money — and it is harder to copy than it looks. No global player is designed from this access point. Bolting it on requires rebuilding the acquisition funnel from zero, against an existing user base that has been trained to expect the standard onboarding pattern. We built the funnel inverted from the start because we built it for the user, not the conversion analytics dashboard.

The illiterate mother holding her child's exercise to the camera, the elderly farmer pointing at a diseased plant, the night-shift driver showing an engine warning light — they do not graduate from "the keyboard era" before reaching the product. We meet them where they are.

Section 2 — The multilingual reality

2.1 The linguistic context of modern Africa

A child in Abidjan learns mathematics in French. She greets her grandmother in a local language. She slips local expressions into her conversation with friends. Her mother sells in the market in French and a local language. Her father negotiates with a foreign supplier in French and Arabic. Her uncle in Lagos texts her in English mixed with regional vernacular. Her cousin in Madrid replies in Spanish.

This is the linguistic reality of contemporary Africa: colonization, globalization, and cross-border commerce have made the continent one of the most multilingual on Earth. The languages of daily business, education, government, and cross-border communication, however, are largely French, English, Spanish, Portuguese, Arabic, and increasingly Chinese.

In Côte d'Ivoire, French is spoken by people who never went to school. In Senegal, French is the language of administration and most of the press. In Cameroon, French and English coexist at the national level. In Nigeria, English is the unifying language. In Mozambique and Angola, Portuguese is universal. In Morocco and Algeria, Arabic and French are both standard. There is no African user, in 2026, who does not speak at least one of these global languages fluently.

This is the linguistic market Déblo serves at launch — alongside global users who share the same multilingual reality (European parents, Latin American freelancers, Middle Eastern professionals, North American diaspora).

2.2 What Déblo speaks fluently today

Through its Vertex AI backbone — Gemini Flash Live for the real-time voice runtime, Gemini frontier models for multimodal reasoning and live camera understanding, open-weight Gemma in the memory and RAG layer, and Gemini Embedding 2 for semantic search across user memories, tasks, and the RAG corpus — Déblo speaks natively and fluently:

An in-house orchestration layer sits underneath, providing automatic failover across Anthropic Claude, OpenAI, and Mistral for resilience and workload specialisation. The primary reasoning path is Google's frontier stack on Vertex AI from day one.

These are the languages African users — and global users — already speak to conduct business, study, government, and family life across borders. The differentiator is not language count. It is the delivery: real-time voice and eyes, on entry-level mobile, on mobile-money rails, with a behavioural register layer that handles mixed-register input naturally.

2.3 The behavioural moat: register handling without correction

When an Ivorian user says "frero, j'ai pas calé l'exo de maths là dêh, tu peux m'aider ?" to a mainstream AI assistant, the model usually understands the words but rebases its response to formal standard French, occasionally with a corrective hint or a switch in tone. The cognitive distance between input and output makes the conversation feel alien.

Déblo is engineered, through the voice prompt system on Gemini Flash Live, to stay in the user's register: it understands the local connective tissue, does not correct, does not moralise, and continues the conversation in standard French with a complicit, never professorial, tone. The output is not the local register — it is French — but the conversational handling of the mixed-register input is what gives Déblo cultural credibility. A user code-switching into any local register gets the same treatment: understood, not corrected, conversation continues smoothly.

This is the defensible behavioural moat at launch. It is real, measurable, and observable in production. It is not a claim of native local-language output.

2.4 The multilingual roadmap

Native voice output across local African languages is on the platform roadmap for 2027–2028. It is not in the launch product.

Building it requires:

1. Corpus collection at scale — conversational, oral, pedagogical data that does not exist on the public web in sufficient volume for any of these languages.
2. Native speaker linguistic review and prosody calibration for each language — slow, expensive, specialised work.
3. Fine-tuning of voice runtimes on these corpora — open-weight Gemma is well-positioned as the open-weight target for this fine-tuning work, with Vertex AI providing the training infrastructure.
4. Regional accent and register synthesis on fine-tuned voice models — calibrated against authentic regional speech, not generic synthetic voices.

This is a multi-year, multi-million-dollar undertaking. It cannot be funded from B2C credit revenue alone. It requires institutional partnership.

The launch product covers French, English, Spanish, Portuguese, Arabic, German, and Chinese — fluently, in real-time voice and eyes. The roadmap is the path from a multilingual-by-Vertex-AI-backbone platform to one that hosts native voice output in local African languages via open-weight Gemma fine-tuning.

2.5 Why this matters beyond fluency — the institutional thesis

Multiple development finance institutions and multilateral programs maintain active budgets for AI inclusion in Africa and digital language preservation. Together these represent the institutional funding pool eligible to support the multi-year roadmap for native voice output across local African languages.

Déblo's local-language roadmap is not a launch claim. It is a fundable institutional thesis — and it is precisely the work for which development finance grants and multilateral cultural programs exist. The product is in production, the team is in Abidjan, the distribution layer is mobile money, the user base is real, and the technical foundation is ready to host fine-tuned open-weight Gemma weights when the funding arrives.

That positioning is what makes Déblo eligible for B2G and DFI streams that mainstream AI labs will not realistically pursue. We are not promising what we have not built. We are positioning the product that already serves Africa's actual linguistic reality today — and inviting institutional partners to fund the next layer.

Section 3 — Market 1: Daily assistance

3.1 The opportunity

Daily assistance is the widest market for a real-time voice & eyes AI platform. The addressable user is anyone who would rather speak, or show through the camera, than type: parents managing family logistics, traders calculating margins, freelancers drafting client emails, families navigating administrative paperwork, professionals dictating notes between meetings.

The user does not need a vertical product. They need a voice & eyes companion that handles whatever comes up: a question, a translation, a draft message, a calculation, a photo to interpret, a document to summarise, a school report to read aloud. The same product, across the day, across the week, across contexts.

3.2 What Déblo delivers in daily assistance

• Real-time voice conversation for any everyday topic.
• Real-time camera vision — point the phone at any document, sign, prescription, invoice, school report.
• Photo OCR for documents, receipts, contracts, letters, administrative paperwork.
• Translation across the seven launch languages, in voice or text.
• Draft writing assistance for emails, messages, CVs, cover letters, proposals.
• Calculation, conversion, basic financial questions for traders and small-business owners.
• Session memory within a conversation, with WhatsApp / SMS / email follow-up summary on request.

3.3 Why this market matters for the platform

Daily assistance is the entry vertical for the broadest possible audience. A parent who uses Déblo to draft a school note becomes a candidate for K-12 tutoring sessions for their child. A trader who uses Déblo to read a contract becomes a candidate for the Pro tier. A diaspora user who translates a letter becomes a candidate for monthly subscription. A mother who uses Déblo Eyes to read her child's school report aloud becomes the most credible distribution channel: she tells the other mothers in her family WhatsApp group.

This is also the market where Déblo is most directly globally usable from day one. The product needs no vertical adaptation to serve a freelance designer in Casablanca, a delivery worker in Lagos, a parent in Brooklyn, or a small-business owner in Lisbon.

3.4 Economics of Market 1

• Freemium tier: welcome credits at signup, free daily credits forever, no card required.
• Pay-as-you-go: top-up from 100 FCFA (~15 US cents) via mobile money or card.
• Pro subscriptions for power users: $5 to$15 per month depending on usage profile.

Section 4 — Market 2: K-12 education

4.1 The opportunity

Africa has 250 million school-age children. A large share faces structural barriers to quality educational support outside school hours — language gap between home and instruction, limited rural tutor access, household budgets a fraction of private tutoring rates, and parental availability constrained by informal-sector working hours. The K-12 opportunity is global beyond Africa — primary and secondary education systems everywhere share variants of the same gaps — but Africa is where the structural problems converge most sharply.

4.2 The five structural problems

1. Language gap. Children learn in a school language different from their home language, with no translation help when concepts are abstract.
2. Tutor accessibility. Private tutors charge 30 000 to 80 000 FCFA per month — equivalent to the median formal-sector salary in many countries.
3. Parental availability. Parents work twelve to fourteen hours per day, often in the informal sector, and cannot supervise homework.
4. Curriculum specificity. Each country has its own school program. Senegalese BAC differs from Beninese BEPC differs from Ghanaian WAEC differs from Ivorian BAC.
5. Hardware reality. Children use shared family phones, often older Android, with intermittent connectivity.

4.3 What Déblo delivers in K-12 education

• Real-time voice tutoring in French and English on Gemini Flash Live, with multilingual support for the other launch languages.
• Real-time camera streaming via Déblo Eyes — the child holds up the exercise, the homework, the textbook page, the test paper; Gemini frontier models on Vertex AI read it in real time and Déblo guides the resolution by voice. Not OCR screenshot upload. Live conversational vision.
• Country-specific curriculum knowledge (Côte d'Ivoire, Senegal, Benin, Togo, Cameroon, DR Congo, Ghana, Morocco at launch; further countries in Phase 2).
• Exam preparation with BAC, BEPC, WAEC, KCSE, GCSE, CEPE, BFEM, DEF, and other regional programs.
• Socratic method by default — Déblo guides through questioning rather than handing over answers.
• Parental session reports via WhatsApp summary on request.

Déblo does not replace teachers. It supports learners outside the classroom, fills the parental-availability gap, and gives families a tool that fits their actual budget and hardware.

4.4 Economics of Market 2

• Freemium tier: welcome credits at signup, free daily credits forever, no card required.
• Pay-as-you-go: top-up from 100 FCFA (~15 US cents) via mobile money.
• School partnerships: per-classroom licensing at 5 000 to 15 000 FCFA per month, paid by school administration.

Real cohort metrics — MAU, conversion, retention, school partnerships — will be published 90 days post-launch (September 2026), grounded in real traction from the June 1, 2026 launch.

Section 5 — Market 3: Customer support

5.1 The opportunity

The African telecom and ISP sector operates approximately 50 000 active level-1 customer service agents across major operators (telcos and ISPs). Average agent fully-loaded cost: 200 000 to 400 000 FCFA per month per seat.

Most level-1 inquiries are repetitive: balance check, top-up issue, package activation, basic troubleshooting, address change, SIM swap. These are the exact patterns real-time voice AI handles well in production today. The opportunity extends to financial services (mobile money support), public utilities, e-commerce, healthcare triage, and any support function with a repetitive level-1 workload.

5.2 The inbound signal

In April 2026, before any outbound effort, a service manager at a Top-5 West African ISP reached out to ZeroSuite unsolicited after experiencing the product. The message indicated that Déblo's voice agent capability could be applied to their level-1 call center operations — a workforce of roughly 500 agents (name available under NDA).

That inbound signal validates the horizontal platform thesis. If one regional ISP recognises the opportunity unprompted, larger operators across the continent will too — and they have orders of magnitude more agents and budget.

5.3 What Déblo delivers in customer support

• Real-time voice agent answering subscriber inquiries in the local language register, with the local accent, on Gemini Flash Live.
• Real-time camera capability — the subscriber shows a SIM card, a router LED status, an error screen, a bill; Déblo identifies the issue in the same call.
• Native handling of mobile money flows (balance, top-up, transfer queries).
• Code-switching handled without escalation when the user mixes French with a local register mid-sentence.
• Integration with telco billing systems via standard APIs.
• Hand-off to human agent when complexity exceeds threshold.
• Per-call cost target: 50 to 100 FCFA per call vs 800 to 1 500 FCFA per call with human agents at the equivalent quality bar.
• Response latency target: phone-call-like response time, equivalent to or better than human agent SLAs.

5.4 Pricing model

• Per-call pricing: 50 to 100 FCFA per resolved call (~10 to 20 US cents).
• Volume tiered: enterprise contracts at 10M+ calls per month with negotiated unit economics.
• Setup fees for integration with billing and CRM systems.

Target Phase 1 deployment: two pilot contracts signed (one telco, one ISP) by end 2026.

5.5 Why the platform fits this market

Mainstream support automation tools are mature in English-speaking Western markets. Few of them handle West African French mixed with local registers. Few of them route mobile money flows natively. Few of them have integrated real-time camera vision for visual troubleshooting. Few of them are accountable on Lagos, Abidjan, or Dakar SLA.

Déblo is built to deploy in this environment without a multi-quarter localization project — and the same platform can be deployed for global support operations where multilingual coverage and natural voice flow are the requirement.

Section 6 — Market 4: Languages and inclusion

6.1 The opportunity

Public sector and development finance buyers maintain two converging budget streams that the platform is structurally eligible for:

1. AI for development digital transformation programs maintained by major development finance institutions, multilateral funders, and bilateral development agencies — collectively, committed budgets for AI in Africa exceed $1.5 billion in active programs as of May 2026 based on public disclosures.
2. Language preservation and inclusion programs maintained by multilateral cultural agencies, regional cultural preservation funds, and country-level cultural ministries. Smaller individual budgets but growing fast and politically prioritised.

The market extends beyond Africa wherever voice access for non-readers, multilingual citizens, or speakers of underserved languages is a public-policy priority.

6.2 What Déblo delivers in this market

• Voice-enabled citizen services (tax form explanations, agricultural extension advice, public health hotlines) in the languages the platform supports today, with a roadmap for local languages.
• Real-time camera vision for users who cannot read — a mother holds up her child's report card, an elder holds up a prescription, a market trader holds up an invoice. Déblo reads aloud and answers questions.
• Adult literacy support for users who cannot read but can speak.
• Cultural and historical knowledge bases accessible by voice.
• Public sector workforce support (teacher assistants, healthcare worker triage tools, civil servant onboarding).
• Documented impact metrics suitable for DFI reporting (Theory of Change, Logical Framework, OECD-DAC criteria).

6.3 Why mobile money distribution unlocks this market

Most development-finance-funded digital tools fail at distribution. They are built remotely, hosted on infrastructure that does not optimise for African network conditions, and paid for in dollars by card. African end users frequently do not have the bank cards, the bandwidth, or the latency tolerance to use them.

Déblo addresses this at the foundation: the major regional mobile money providers are in the product across the 6 launch countries, not an afterthought. A grant program can sponsor 10 000 student accounts with a single mobile money disbursement. A government can subsidise call minutes for rural users without a bank rail. This is grant-deployable infrastructure.

6.4 Pricing model

• Grant-funded sponsored deployments: $30 000 sponsors 5 000 children for 6 months of unlimited use (canonical reference point).
• Government licensing per-citizen-served, billed quarterly.
• Custom language pack development: ~$50 000 to ~$150 000 per language, fully owned by the funder.
• Impact measurement reporting bundled (academic partner audits).

Target Phase 1 deployment: one signed DFI partnership and one government pilot by end 2026.

Section 7 — Stack, infrastructure, and mobile money distribution

7.1 The technical stack — Vertex AI native

Déblo runs on Google's frontier stack on Google Cloud Vertex AI from day one.

Multimodal during voice and eyes. Photo OCR (in-house preprocessing pipeline), PDF reading, web search, WhatsApp and SMS messaging, bug reporting, and live camera streaming. All accessible mid-conversation without breaking the voice loop. The combination — voice + live camera + reasoning + web + messaging in one continuous conversation, on Vertex AI — is uncommon among consumer voice AI products today.

7.1.b In-house engineering — built in Abidjan, accelerated by Claude Code

Déblo is engineered in-house from Abidjan by Juste Gnimavo, founder of ZeroSuite, with Claude Code (Anthropic) as senior-engineer-level AI coding partner. This is the velocity that AI-native engineering tooling unlocks: a one-founder team shipping a multi-product real-time voice and camera platform — across K-12, Pro, customer support, and language inclusion — at a pace traditionally reserved for fifteen-person AI startups with $5–10M of seed capital.

The product is not a chatbot wrapper. It is:

• A proprietary orchestration layer over a frontier model bundle — Google Vertex AI (primary), Anthropic Claude, OpenAI, and Mistral — with workload-tuned model selection and automatic multi-vendor failover.
• An in-house preprocessing pipeline validated against an African document corpus across 5 structured testing sessions, with 9 frontier models benchmarked head-to-head.
• A live-camera understanding layer running mid-conversation through the voice loop, on Vertex AI.

Anthropic powers two layers of the stack: Claude Opus on the voice-runtime failover path (guaranteeing continuity when Vertex regional capacity is constrained), and Claude Code as the engineering accelerator without which deblo.ai would not exist.

7.2 The mobile money distribution layer

Déblo integrates natively with the major regional mobile money providers covering the 6 launch countries, plus Stripe for international card payments and selected African card networks for global users.

Entry tier: 100 FCFA (~15 US cents). Pro tier subscriptions for adults: $5 to$15 per month, payable by mobile money or card.

This is the structural decision that makes the platform usable by hundreds of millions of African adults who do not have an international bank card — and the same decision keeps the global card path open for users who do.

7.3 Why this stack is hard to replicate from outside the continent

A large global lab pivoting to the African market in 2027 would need to:

1. License real-time voice & eyes infrastructure regionalized to African network conditions (12 months minimum).
2. Integrate with multiple mobile money operators across the continent (6 to 12 months per operator, often gated by regulatory KYC).
3. Establish on-ground sales for B2B telco contracts (3 to 5 years to build relationships).
4. Earn cultural authenticity that survives marketing scrutiny — built by an African founder, in Africa, against African user input.
5. Calibrate behavioural register handling across multiple regional contexts — staying naturally with the user's mixed input without correcting or moralising — across the regional registers our launch beachhead uses.
6. (When the institutional local-language work is funded and shipped) Build local-language corpora and fine-tune voice runtimes — a multi-year, multi-million-dollar undertaking the platform is positioned to receive via Market 4 partnerships, with open-weight Gemma as the fine-tuning target.

Déblo has a meaningful head start on points 1 through 5. Point 6 is the multi-year roadmap, fundable through DFI partnerships, with the distribution infrastructure already in place to absorb it.

7.4 Why Vertex AI specifically

The choice of Vertex AI as the production backbone is deliberate:

• Latency — Gemini Flash Live delivers consistent phone-call-like latency for voice in production today, including across African 4G networks via European edge regions.
• Multimodal — Gemini frontier models handle voice + image + camera streaming in a single model call, removing the pipeline complexity that other multi-vendor stacks impose.
• Cost — Vertex AI committed-use pricing makes K-12 free tier subsidisation economically viable at scale.
• Reach — Google Cloud's strategic AI growth priority in Africa aligns with Déblo's distribution footprint, opening the door to joint deployment, technical partnership, and open-weight Gemma reference deployment opportunities.
• Open weights option — Open-weight Gemma in the RAG layer provides a fine-tuning target for the future local-language work that closed-source-only stacks would not support.

The in-house orchestration layer sits underneath — graceful degradation across Anthropic Claude, OpenAI, and Mistral when Vertex regional capacity is constrained — but the primary reasoning path is Google's frontier stack from day one.

Section 8 — Deployment plan

8.1 Phase 1 (Q2–Q4 2026) — Six countries, four markets seeded

Countries: Côte d'Ivoire, Senegal, Benin, Togo, Cameroon, Ghana.

• Market 1 (Daily assistance): broadest user base, monetised through credits and Pro subscriptions.
• Market 2 (K-12 education): launched June 1, 2026.
• Market 3 (Customer support): pilot contracts in development.
• Market 4 (Languages and inclusion): institutional partnerships in development.

Launch date: June 1, 2026. First 90-day traction report — cohort metrics, conversion, contracts pipeline — published September 2026. The targets above are pre-launch state.

8.2 Phase 2 (2027) — Twelve additional countries

Francophone: Mali, Burkina Faso, Niger, DR Congo, Gabon, Madagascar. Anglophone: Kenya, Nigeria, Tanzania, Uganda, Rwanda, South Africa.

• Market 1: continental rollout following mobile money integration coverage.
• Market 2: expanded K-12 footprint.
• Market 3: expanded enterprise contracts.
• Market 4: expanded development-finance partnerships.

Deployment milestones across the four markets — and the corresponding cohort metrics — will be published 90 days post-launch.

8.3 Multi-year targets

Detailed multi-year targets across all four markets will be published 90 days post-launch (September 2026), once first-cohort traction data is available. The model is shared with qualified investors under NDA pre-launch.

Three reference scenarios for the conservative B2C funnel (no B2B telco upside included):

Math: Year 1 = 100 K × 5 % × 1 500 FCFA × 12 ≈ 90 M FCFA ≈ $150 K. Year 3 = 2 M × 9$9 M. Mix at maturity targets ~60 % B2C micro top-ups, ~25 % Pro, ~15 % B2B telco.

Pre-launch direction: revenue mix at scale targets a balanced split across B2C credits and subscriptions, B2B and B2G contracts, and institutional grant-funded deployments. The free tier remains free, sustainably, funded by the paying minority and the institutional layer.

8.4 Go-to-market — the Coca-Cola playbook

We will not out-engineer the giants. We will out-distribute them.

The urban-literate top decile already uses ChatGPT, Claude, Gemini — in San Francisco, in Paris, in Abidjan's Plateau. The other 90 % does not — not because the AI is weak, but because nobody runs vernacular ad campaigns in Abidjan, nobody trains village ambassadors, nobody hands tablets to rural classrooms. On a VPN connection from London or Paris, AI ads run every 3 minutes on YouTube and Facebook. On a Côte d'Ivoire IP, the inventory is empty. The same pattern is observable in rural France, the Mississippi delta, the Brazilian Nordeste, and the Indian villages outside the major metro areas. The playbook for reaching these markets is Coca-Cola, not Silicon Valley — every village, every language, every screen.

Four parallel distribution levers post-launch:

1. Vernacular ambassadors. On-the-ground sales reps trained per launch country, speaking Wolof, Bambara, Dioula, Lingala, Hausa, Yoruba, Swahili, and Pular. The neighbor who explains Déblo in the user's own language is a distribution channel that does not appear in any global AI lab's go-to-market plan.
2. NGO + humanitarian + education partnerships. Literacy associations, primary schools, vocational centers, humanitarian NGOs — institutions that already have the trust and the field presence we cannot build alone in 12 months. The same channel routes the institutional grant-funded deployments disbursed via mobile money described in Market 4.
3. Devices in classrooms. Tablets, phones, and mini-PCs placed in classrooms and youth centers with Déblo at the center — group tutoring, voice + camera, French / English / local language. Sales motion seeded by founder + 2 field collaborators recruiting.
4. Vernacular ad buys. Targeted YouTube, TikTok, Facebook, and WhatsApp Status campaigns on African and diaspora geos — the inventory the giants are not buying. Cost-per-install a fraction of Western markets, with conversion rates supported by the zero-onboarding moat described in Section 1.5.

We will not win the benchmark race against OpenAI, Anthropic, or Google. We will win the last-kilometer race they decided not to run. The 90 % the giants abandoned is not a niche — it is the majority of humanity.

Section 9 — Why this works now

Five macro factors converge in 2026:

1. Voice & eyes AI maturity. Real-time voice runtimes reached production-grade quality in late 2025. Real-time camera vision via Gemini Flash Live reached production-grade quality in early 2026. Phone-call-like latency paired with live camera understanding is now feasible on entry-level hardware.
2. Mobile money saturation. Wave, MTN Mobile Money, and Orange Money together cover the majority of digital payments in francophone West Africa. The card-payment barrier is gone for the launch beachhead.
3. Smartphone penetration. Africa crossed 50% smartphone penetration in 2024, projected 80% by 2030. Even budget Android can run Déblo.
4. AI inference cost decline. High-quality LLM inference dropped roughly 90% in cost since 2023. Pricing at ~15 US cents per session is economically viable at scale.
5. Demographic momentum. Africa: 250M students today, 450M projected 2050 — one of the largest educational markets globally, structurally underserved. Combined with the global voice-first user base across other regions, the macro tailwind is in place.

These factors align uniquely now. Three years ago, the platform was technologically infeasible. Three years from now, the market will be competitive — but Déblo will already have the brand, the data, the cultural authenticity, the mobile money integrations, the Vertex AI committed-use position, and the customer relationships.

Section 10 — Call to action

Investors and funders

Bootstrap to date. Running a seed round. Specific terms shared on request to qualified investors. Use of funds: ~40% engineering and infrastructure (Vertex AI commit, Gemini token budget, multimodal hardening), ~30% go-to-market (B2B telco sales, DFI grant matching, school partnerships, diaspora paid acquisition), ~20% operations and compliance, ~10% strategic reserve. Target runway: 18–24 months. → [email protected]

Foundations, NGOs, and multilateral institutions

Sponsored deployments for rural and underprivileged regions. $30K USD sponsors 5 000 children for six months. Documented impact, audited by academic partners. → [email protected]

Technology partners (Google Cloud, voice infrastructure, mobile money rails)

Joint case studies, reference deployments, Vertex AI committed-use partnerships, open-weight Gemma reference deployment, technical roadmap collaboration. → [email protected]

Enterprise customers

Telcos, ISPs, financial services, healthcare, public utilities operating customer support at scale. Pilot deployments available. → [email protected]

Media and journalists

African solo founder shipping production-grade real-time voice & eyes AI from Abidjan on Vertex AI, with a small team and orchestrated AI agents handling implementation. Field tour access available. → [email protected]

Regulatory and privacy posture

Déblo operates under Côte d'Ivoire's Loi 2013-450 framework for personal data protection, the UEMOA regional data protection directive, and GDPR for diaspora users in the EU. Mobile money payment flows comply with operator-specific KYC requirements across the 6 launch countries. Data residency, retention, minor consent handling (critical for the K-12 vertical), and audit trails are designed into the product architecture, not bolted on. A dedicated compliance brief covering data flows, access controls, retention windows, and incident response is available to enterprise, B2G, and DFI prospects on request. → [email protected]

Conclusion

Déblo is a real-time voice & eyes AI platform deployed on Google Cloud Vertex AI. The user speaks. Déblo answers live, like a phone call. The user shows their world through the camera. Déblo sees what they see, in real time, and the conversation continues. The platform is built for people who prefer speaking and showing over typing — and that preference is shared by hundreds of millions of people inside Africa and beyond.

Africa is the launch beachhead, the cultural anchor, and the strongest initial market. It is not the ceiling. The same product, with the same Vertex AI foundation, serves daily assistance, K-12 education, customer support, and language inclusion — across geographies, currencies, and registers, reaching toward the 1 billion adults the global AI giants have not designed for.

Created in Abidjan. Built for the world. Deployed on Vertex AI.

Knowledge accessible to anyone who can speak — or show.

Bringing access to expertise — to 1 billion people who never had any.

If you operate, fund, or report on AI infrastructure, voice-first products, or development finance work, you are a potential partner. We invite you to reach out.

Contact

Juste Azandegbe Gnimavo Founder & CEO, ZeroSuite Inc. [email protected] Abidjan, Côte d'Ivoire

Web https://deblo.ai (product) https://pulse.deblo.ai (live metrics dashboard) https://zerosuite.dev (parent company) https://justegnimavo.com (founder personal site) https://thalesandhisaictoclaude.com (engineering blog — 1 100+ articles, EN / FR / ES)

Document license This white paper is published under Creative Commons Attribution 4.0 (CC BY 4.0). It may be freely shared, quoted, and translated, with attribution to ZeroSuite Inc.

Document version history

• Version 1.0 — April 2026 — Initial publication (education-only frame, archived in git history).
• Version 2.0 — May 2026 — Real-time voice AI platform thesis aligned to the four-market foundation.
• Version 3.0 — May 2026 — Real-time voice & eyes AI; Vertex AI / Gemini / open-weight Gemma backbone made explicit; in-house orchestration and Anthropic partnership documented.
• Version 3.1 — May 2026 — Added Section 1.5 (accessibility moat / zero onboarding), Section 8.3 reference revenue scenarios, Section 8.4 (Go-to-market — Coca-Cola playbook). Updated thesis around the "1 billion people" framing. Current document.
• Version 4.0 — Planned Q4 2026 — Updated with first six months of deployment data.

End of white paper.