Production platforms and AI systems shaped around the constraints of the teams and people who use them.
Selected projects
Full Stack
REWA Digital Rent & Rewards Platform
Cross platform architecture for the UAE's digital rent and rewards platform, connecting a Flutter tenant app, Next.js landlord portal, Symfony operations console, and NestJS services across payments, onboarding, leases, rewards, notifications, and identity workflows.
Tenant mobile, landlord, and operations experiences share consistent contracts through a central service layer
Payment orchestration supports cards, bank transfers, direct debit, consent, reminders, and collection exception lifecycles
UAE Pass, Ejari, invitation, lease, and identity journeys are modeled as resumable cross channel workflows
Rewards, vouchers, notifications, and rent records flow through modular services, workers, and a shared schema authority
A self organizing knowledge base that uses autonomous agents to crawl, classify, and link enterprise data sources. Features multi agent orchestration for continuous graph updates, consistency verification, and gap detection across distributed knowledge stores.
Self healing knowledge graph with automatic inconsistency detection and correction
Autonomous research agents that proactively identify and fill knowledge gaps
Hybrid vector graph retrieval enabling complex multi hop reasoning queries
Enterprise grade RBAC with granular collection level permissioning
The core execution engine enabling AI agents to interact with modern web applications reliably. Handles complex DOM interactions, authentication flows, and dynamic content extraction through semantic understanding and anti detection browser orchestration.
Headless browser orchestration with anti detection and fingerprint randomization
Semantic DOM understanding for resilient, self healing selector generation
High performance async runtime built on Rust and Tokio
Plugin architecture supporting custom action handlers and middleware
A high performance vector search platform engineered for billion scale embeddings. Implements custom HNSW indexing strategies with distributed sharding for sub millisecond query latency across multi modal embedding types.
Distributed index sharding with automatic rebalancing and failover
Real time vector insertion with tunable consistency guarantees
Multi modal embedding support across text, image, and audio
Production grade observability with per collection Prometheus metrics
A real time ad fraud detection system using autonomous agents to simulate user behavior and validate traffic sources. Processes 100K+ events per second with deep learning behavioral analysis and automated threat rule generation.
Real time behavioral analysis using deep learning classification models
Autonomous traffic validation agents simulating genuine user patterns
High throughput event ingestion pipeline handling 100K+ events per second
Automated rule generation that adapts to emerging fraud patterns in real time
Omair Afzal