Architected and launched a real-time Digital Twin platform for fleet telemetry, anomaly detection, and mission replay. Owned product strategy, system architecture, API design, frontend development, edge deployment, and observability. Built a distributed edge-cloud platform capable of processing high-frequency telemetry streams with real-time operational insights.
Production-ready AI KYC and compliance automation for regulated teams. An agentic workflow that runs customer onboarding, document verification, and AML risk assessment end to end: six LangGraph nodes, seven mechanically-enforced invariants, and a guardrail before every human review. Cuts a 3.5-hour manual dossier to 14 minutes (93% faster document review), deployed EU-resident on Fly.io Amsterdam for GDPR-aligned data residency. Built to hand an auditor a clean trail, not a demo.
Earlier product work. Exploratory projects that informed how I now ship to production.
NeuroPro
Production-ready on-device AI for iPhone: real-time computer vision rendered entirely on the Metal 3.0 GPU and Apple Neural Engine. Object detection, scene classification, style transfer, and segmentation at 1080p 60fps with an 11.1ms preview frame time and zero dropped frames. Every pixel processed on-device, no network permission, no analytics, no PII. Privacy-first edge AI that respects the user.
11.1 mspreview frame time
60 fps1080p capture
On-deviceANE + Metal 3.0
Swift + Metal 3.0Vision + Core ML + ANEiOS + GPU Compute Shaders
Live at books.dalecabra.com: a production book recommendation system that blends content-based and collaborative filtering with explainable recommendations. An AI reading companion matches mood and taste across a real catalog, with Redis caching, model monitoring, drift detection, and a rate-limited API behind a React + TypeScript frontend. Every recommendation is explainable, not a black box.
Architected and shipped a native iOS pet wellness app in Swift, enabling owners to monitor health trends, track behavior patterns, and receive AI-powered personalized care recommendations. Owned product vision, iOS architecture, Core Data persistence, and integration with the AI analytics backend from concept to deployment.
AI that gets surfaced in the dashboard. Shipped a Looker suite for JerseySTEM, a US nonprofit running on Salesforce + Looker, that turns raw fundraising operations into one source of truth: incoming anniversaries, active sponsors, newsroom content, and a per-sponsor view the relationship team opens every Monday. RAG pipelines and agents clean and enrich the underlying data at write time; a governed LookML semantic layer keeps every metric defined once. The development director sees the same number the ED sees; the board sees the same number the auditor sees. One artifact everyone trusts, built so the volunteer tech team can extend it in code after I left.
Active research. Building the next generation of AI infrastructure.
AI News Dashboard
Built a context-aware AI news platform that adapts UI and content to user behavior, session history, and environmental signals. Delivered Arc-style search with contextual suggestions, an AI skill orchestrator for summarizing and comparing content, and real-time social collaboration. Achieved sub-2.5s load times through advanced caching and edge optimization.
Built a social funding and prediction market platform for startup founders, combining real-time prediction markets with SEC-compliant fundraising tools. Architected binary and categorical market creation with live price updates, plus Regulation CF and Rule 506(c) compliant investor onboarding with KYC/KYB workflows and accreditation tracking.
