FarmLens

AI-Powered Farm Management, Learning & Extension Support
for Kenyan Smallholder Farmers

UNDP Kenya — timbuktoo AI Lab AI Discovery & Maturity Assessment August 2026

The Problem We Solve

02 / 12

The Extension Gap

Most small-scale farmers across Sub-Saharan Africa rely on guesswork when deciding what to plant, when to plant, or how to manage crop and pest threats. Climate change continues to disrupt traditional farming practices, leading to:

Crop failure and reduced yields
Wasted inputs (fertilizer, pesticide, seed)
Financial losses and food insecurity
Limited access to qualified agronomic advice

The Digital Divide

Kenyan smallholders face additional barriers to technology adoption:

Unreliable internet connectivity in rural areas
Low digital literacy among older farmers
Language barriers — English-only tools exclude Kiswahili & vernacular speakers
Fragmented tools — no single platform for records, learning, and support

FarmLens bridges this gap with an integrated, offline-aware, multilingual platform designed specifically for Kenyan smallholder contexts.

What FarmLens Does

03 / 12

A mobile-first platform that connects everyday farm records with reviewed learning, reminders, guided questions, human extension support, and organization workspaces.

Farm Records

Activities & field notes
Income & expense tracking
Harvest quantities
Input purchases & usage
Photo galleries per farm

Learning & Assessment

Published, reviewed lessons
English & Kiswahili content
5-area skills assessment
Personalized lesson ordering
Flashcards & knowledge checks

Ask FarmLens (AI)

Voice-first Q&A interface
Grounded in published content
Saved conversation threads
Safety classifier & fallback
Human escalation for high-risk queries

Reminders

Create & update farm tasks
Complete, snooze, reschedule
Restore dismissed reminders

Human Support

Open support requests
Case timeline & messaging
Staff replies & appointments
Call, office, or field visits

Cooperatives

Member context view
Produce delivery capture
Cooperative announcements
Privacy-respecting access

The AI Solution: Ask FarmLens

04 / 12

Core Task

Ask FarmLens is a retrieval-augmented generation (RAG) assistant that answers farmer questions using only published, reviewed FarmLens learning content. It supports text input and, on compatible devices, browser-based speech recognition and spoken output.

Measurable Outcomes

↑ Farmer knowledge & self-service

↑ Learning content engagement

↓ Routine query load on extension officers

↑ Better farm-level decision-making

Safety & Boundaries

Grounded only in published, approved learning content — no hallucinated agronomic advice
Safety classifier intercepts high-risk wording (chemical dosing, emergency animal health, financial advice)
Human fallback — high-risk queries are escalated to extension officers, not answered by AI
Explicit disclaimers — AI output is general guidance only; never replaces professionals or product labels
Traceability — every response stores provider, model, source identifiers, and response time

AI is currently disabled by default. The mock provider is for development only. Live provider activation is gated on governance, safety, and quality evaluation.

Voice AI & Multilingual Focus

05 / 12

104

Farmers Tested (Kakamega)

5

Cooperatives (Mumias Sub-county)

2

Pipeline Pilots (Kabete)

3

Languages (EN / SW / Kikuyu)

Completed

Assumption Validation & Voice AI Testing

Initial testing phase with 104 farmers across 5 cooperatives in Kakamega (Mumias Sub-county). Focused on validating core product assumptions and testing our Voice AI interface in Kiswahili directly with smallholders in field conditions.

104 Farmers

5 Cooperatives

Kiswahili Voice AI

Upcoming

Pipeline Pilot — Kabete Sub-county

Expanding multilingual coverage into Kabete Sub-county, where two farmer cooperatives have agreed to pilot our multi-lingual Voice AI (specifically testing Kikuyu language integration). They are committed to walking the development journey with us as pilot partners.

2 Cooperatives

Kikuyu Voice AI

Kabete Sub-county

Technical Architecture

06 / 12

Farmer UI
Blade + Tailwind + Alpine

→

Application Service
Laravel 13 / PHP 8.3

→

Eloquent / Repository
Owner-scoped queries

→

PostgreSQL
Production Target

NativePHP Mobile
Android (v3.3.x)

→

Sanctum API
Scoped tokens

→

Sync Service
ULID + conflict preservation

→

SQLite (Local)
Offline persistence

AI Provider Layer

Provider-neutral interface — supports disabled, mock, and OpenAI-compatible HTTP adapters
Currently configured for Google Gemini integration
Swappable adapters for future providers
Queued application services for production scale
Retrieval returns only published content with approved translations

Security & Auth

Web: Breeze CSRF/session protection
Native: Sanctum scoped tokens + API throttling
Spatie Permission for role-based access
Owner-scoped queries for all farmer data
Audit records for sensitive changes
Rate limiting on AI, sync, and auth endpoints

Data & Content Governance

07 / 12

Learning Content (AI Corpus)

Internally authored agricultural lessons, assessments, flashcards, and knowledge checks
Translated into English and Kiswahili
Governed through a draft → review → approve → publish workflow
Only published content with an approved translation is visible to farmers or used by AI retrieval
Content and translations version independently

All demo seed content is explicitly labelled and must not be treated as validated agronomic advice.

Conversation Data

Question text, response text, provider identifier, model identifier
Source content identifiers for traceability
State and response time for operational monitoring
Stored per-user for thread history, rename, and delete controls

Private farmer records (financial data, personal details) are NEVER used as AI context. They are stored separately under owner-scoped authorization.

Content Review Workflow

Draft

→

Review

→

Approve

→

Publish

→

Withdraw (if needed)

Responsible AI & Ethics

08 / 12

Safety-First Design

AI output is explicitly framed as general guidance only. It does not replace extension officers, agronomists, veterinarians, emergency services, or registered product labels.

High-Risk Escalation

Safety classifier and keyword checks intercept queries about chemical dosing, emergency animal health, and financial advice — routing them to human extension officers.

Content Boundaries

Defined boundaries for chemical, veterinary, financial, and emergency guidance. Content reviewers check agricultural accuracy, language, safety wording, and source quality before publication.

Data Minimization

The AI system does not process sensitive personal data (ID numbers, medical info, etc.). Farm records are stored separately and are owner-scoped; never used as AI context.

Transparency & Traceability

Every response stores provider, model, and source identifiers. Farmers can see traceable sources and content versions where appropriate.

Legal & Regulatory

Privacy Policy and Terms of Use exist but require final Kenya-specific legal approval. Operational privacy workflows (consent, export, deletion, retention) are in progress. DPIA planned before production launch.

Offline-Aware Synchronization

09 / 12

How It Works

Clients persist records and operations before reporting success
POST /api/v1/sync/push accepts up to 100 operations and deduplicates by operation ULID
Supported records: activities, harvests, and financial transactions
Server detects stale updates; conflicts return authoritative server data without deleting local data
GET /pull?since= returns owner-scoped changes
GET /status returns pending counts

Append-oriented financial records should be corrected with a new entry. Profile conflict merging remains later-phase work.

NativePHP Mobile

Pins nativephp/mobile 3.3.x
Android shell with embedded PHP 8.4.23
Local SQLite database for offline persistence
Pending sync_operations created in offline mode
Automated tests verify persistence and queue creation

Physical-device testing still required for: login across process death, secure token storage, background reconnect, push retry/idempotency, and conflict display.

The FarmLens Team

10 / 12

Clinton Adoga

CEO & Co-Founder

25-year-old Kenyan social entrepreneur and climate-tech innovator. UNITAR Certified Climate Youth Champion, Mercedes-Benz BeVisioneers Fellow, YALI East Africa Alumnus, and TEF Entrepreneurship Programme Fellow. Leads vision, partnerships, and strategic growth.

Stella Njambi

Co-Founder & Technical Lead

Laid the technical foundation for FarmLens. Leads architecture, Laravel application development, NativePHP mobile build, provider-neutral AI architecture, and offline synchronization framework. Drives feature development and platform engineering.

Lewis Kamau

Co-Founder & Technical Contributor

Works closely with Stella to fine-tune the user experience and add new features. Supports technical development, AI integration, mobile build pipeline, and platform hardening for production readiness.

Extended Team & Advisors

Supported by content reviewers (agricultural accuracy), extension officers (support workflow design), and cooperative officers (user context). Youth field agents are trained as digital support contacts in pilot communities.

Current Status & Verification

11 / 12

55

Tests Passing

160

Assertions

5

User Roles

2

Languages (EN / SW)

What's Working

Responsive desktop and mobile navigation
Progressive onboarding (language, farm, crop)
Multiple farms, crop cycles, records, reminders
Published learning with approved translations
Extension queue, cooperative operations, content governance
Role and permission separation
ULID-based sync with conflict preservation
Provider-neutral AI contracts

Pre-Production Boundaries

AI is disabled by default — mock provider only
Voice uses browser SpeechRecognition (device-dependent)
No managed PDF/video asset pipeline yet
Some staff forms still need drawer conversion
Some copy remains hard-coded in English
Production object storage not configured
Privacy export/deletion workflows incomplete
Automated background sync needs device validation

Status: Functional pre-production platform. Suitable for controlled demonstrations and pilot preparation. Not yet approved for unrestricted public use.

How timbuktoo Can Support FarmLens

12 / 12

We are currently utilizing AWS cloud credits for infrastructure. The timbuktoo AI Lab could provide the most valuable support in these areas:

1

AI Provider Evaluation

Structured support to evaluate real providers (beyond Gemini) for retrieval quality, Swahili language accuracy, safety failure modes, and escalation behavior.

2

Compute for Embeddings

Compute resources to generate embeddings for our agricultural content library or experiment with fine-tuned open models for offline or low-bandwidth scenarios.

3

Speech & Voice Services

Evaluation of server-side speech-to-text and text-to-speech services for consistent Kiswahili and Kikuyu language support, replacing browser-dependent voice.

4

Production MLOps

Technical assistance on monitoring, prompt versioning, A/B testing, and safe rollout of AI features to our pilot users across Kakamega and Kabete.

5

Data & AI Safety Review

External review of our safety classifier, content boundaries, and RAG pipeline to ensure we meet responsible AI standards for agricultural advice.

6

Training & Capacity Building

Compute and technical training for our youth-led engineering team to build MLOps competency and sustainable AI operations.

Ready for the AI Discovery & Maturity Assessment

Contact: Stella, Lewis & Clinton | FarmLens LTD | app.farmlens.africa