What We Can Build For You
Every capability listed here exists in production across our own portfolio. We build nothing we haven't already shipped, scaled, and operated ourselves.
Start a conversation →Maktabi's outreach platform is not a chatbot. It is a layered, self-improving AI system with 208 production service components orchestrating every inbound message in real time. We can embed this architecture into any B2B or B2C product.
Three AI models work in parallel per message: a fast reasoning model (GPT-4o-mini) for intent and sentiment classification, a deep reasoning model (DeepSeek-R1) for strategy and qualification, and a creative model (GPT-4o) for empathetic, context-aware response generation. Each turn runs the right model at the right cost.
Every lead is profiled in real time — their role (owner, agent, accountant, decision-maker), detected persona type (skeptic, trust-seeker, risk-averse, early adopter), language (Arabic, English, Urdu, Tagalog — AI-detected, not rule-based), emotional state (derived from the last 5 sentiment readings), and emotional intensity on a 1–10 scale. Responses adapt accordingly.
Every conversation exists in one of 27 governed states — nurture, soft_close, recover, retain, close_now, disengage, deal_close, agency_convert, churn_prevent, and more. State transitions follow strict rules enforced by a dedicated governance service (no "vibe-based" decisions). Every transition is logged with timestamp, reason, and prior state history.
A feature-vector-based prediction model computes conversion probability, churn probability, upsell probability, appointment probability, and discount interest per turn — cached for 5 minutes, invalidated on signal change. A/B testing variants are assigned per conversation, and outcomes feed back into strategy weights via the Adaptive Learning Engine.
14 objection categories (price, timing, trust, need, competitor, authority, complexity, features, data privacy, and more) each have dedicated rebuttal strategies. Strategies load adaptive weights from the real-time optimisation engine — if a tactic performs poorly, its weight decreases automatically and a softer approach is preferred. The system learns what works on your audience.
The Autonomous Revenue Maximisation Engine runs on every inbound: upgrade probability prediction, revenue risk scoring, urgency classification, opportunity detection, discount decision logic (with sentiment and sentiment guards — no discount offered if emotional_state is anger), and CTA selection. Discounts are one-use per conversation and require minimum onboarding progress before activation.
Intent detection, urgency classification, objection bypass generation, pitch generation, and directive decision run in sequence per message. Directives include: accelerate, cta, micro_pitch, urgency_push, objection_bypass — each generating a different tone and CTA. The system knows when to push and when to wait.
Leads who send audio notes have their messages transcribed using OpenAI Whisper, cleaned by a bilingual cleanup service, and fed into the same orchestration pipeline. Images (receipts, documents, property photos) are analysed by GPT-4o Vision and classified — enabling fully autonomous document-based conversations without human review.
Six services work together to build a real-time lead profile: IndustrySegmentationEngine (classifying as agency, property_manager, landlord, tenant, or general_consumer), QualificationFieldExtractor (location, budget range, number of units, rental frequency, property type, urgency, timeline), QualificationGapDetector, SmartQuestionGenerator, RealTimeQualificationEngine, and QualificationFlowDirector. The pipeline knows what it has and what it still needs — and asks for it naturally, one field at a time.
Dedicated engines handle the post-conversion lifecycle: ChurnPredictionEngine (predicts early, mid, and late-stage churn risk), ChurnPreventionOrchestrator (intervenes proactively before loss), RenewalCampaignService and RenewalTimingOptimizer (maximise renewal conversion at the right window), and ReferralDetectionService (identifies satisfied customers who are likely to refer and triggers a referral outreach sequence automatically).
Before any message is sent: VerificationAgent checks it for factual accuracy, PersonaAdaptationAgent adjusts tone to the detected persona, FeatureRenderer intercepts feature-inquiry intents and renders truth-locked responses (preventing the LLM from inventing features), and RepetitionHandlingService ensures no message pattern repeats within a conversation. Every outbound is audited before delivery.
The AdaptiveLearningEngine records PerformanceSignals for every strategy outcome (positive: conversion, appointment scheduled, objection resolved; negative: churn, escalation, unsubscribe). Strategy weights are recalibrated daily. After 30 days, the system's conversion strategy is tuned to your specific audience — not a generic dataset.
Every product we've built runs in production for paying customers. We know what breaks at scale and have already fixed it.
Full data isolation per organisation — tested against cross-tenant leakage. Every query scoped at the model level. Filament admin panel with per-tenant context enforcement.
Production-grade inbound and outbound webhook pipelines with retry logic, signature verification, queue-based processing, and dead-letter handling.
Tap payment gateway integration with settlement reconciliation, VAT-compliant receipt generation, ZATCA e-invoicing, and subscription billing lifecycle management.
Laravel Horizon-managed queue infrastructure with priority lanes for time-sensitive jobs (payment confirmations, message dispatch) and background lanes for analytics and learning cycles.
Full 360dialog integration — inbound routing, template management and approval monitoring, 24-hour session management, voice message ingestion, media download, and broadcast campaign tooling.
Arabic-first design system with automatic RTL layout switching, Hijri date handling, Saudi phone normalisation, Arabic number detection, and locale-aware content rendering.
Every state change, decision, transition, and AI output is logged with structured context. Full replay capability for debugging any conversation or transaction at any point in history.
Flutter-based cross-platform mobile apps (iOS & Android) with offline capability, deep push notification integration, and shared Dart codebase for rapid iteration across platforms.
Three products, one doctrine: SATE v2 (language-agnostic L0–L4 certification), SATE-STE (JS/TS test generation, healing, mutation, regression), and SATE-Laravel (Laravel causality auditing, Green Lies prevention, blockchain-signed evidence).
We take on select external engagements when the problem is interesting enough and the partnership is the right fit.
End-to-end WhatsApp or web-based AI representive systems — from persona definition and intent taxonomy to full orchestration, A/B testing, and adaptive learning. Deployed on your infrastructure, tuned to your domain.
If you have a domain-specific problem that warrants its own SaaS product — we design, architect, build, and launch it. We've done this in property, HR, procurement, and software auditing. Talk to us about your sector.
Already have a product? We can embed a qualification pipeline, prediction engine, objection handler, or revenue optimiser into your existing flow — as a modular addition, not a rewrite.
Tap integration, ZATCA e-invoice compliance, subscription lifecycle management, dunning automation, and settlement reconciliation — built to Saudi regulatory standards from day one.
A family of three purpose-built products: SATE v2 for language-agnostic execution certification (no tests required, pure observation, L0–L4 layered model); SATE-STE for autonomous JavaScript/TypeScript test generation, mutation testing, healing, and regression detection; and SATE-Laravel for Laravel-specific causality auditing, Green Lies prevention, and blockchain-anchored forensic evidence. Each product operates strictly outside your toolchain — they observe, they never modify.
Acquiring, partnering with, or investing in a technology company in the region? We can independently audit the system's architecture, codebase quality, and operational claims before you commit.
Ready to build?
Engagements start with a conversation, not a proposal. We scope precisely before committing anyone's time.
Request a Quote →