Senior AI Systems & Backend Engineer building production software across AI, distributed systems, cloud infrastructure, and privacy-first applications. From architecture and implementation to deployment, evaluation, and operation.
15+ years building software. Recently focused on production AI.
Software designed, built, and shipped for real users.
Human-in-the-loop document approval for AI agents. Agents submit documents over REST or a hosted MCP server; humans approve, reject, or send them back with comments anchored to the text. Every version is immutable and SHA-256 hashed, and every decision lands in an append-only, hash-chained audit log.
Web · REST API · Hosted MCP server
Next.js · REST · MCP · Scoped API keys · Hash-chained audit log
AI-powered children’s storytelling app for iPhone and iPad. Built around Apple Intelligence and on-device generation, with privacy as a core architectural constraint — stories are generated locally, with no accounts and no data collection.
iPhone · iPad · App Store
SwiftUI · Apple Intelligence · Foundation Models · SwiftData
A private AI research workspace for macOS. NotebookLM-style Q&A runs fully on-device: ingestion, chunking, embeddings, vector storage, and RAG inference all run locally on Apple Silicon with MLX. No cloud, no accounts, no document content leaving the machine.
macOS · Apple Silicon
Swift · MLX · Local LLMs · RAG · Embeddings · Vector store
Privacy-first iOS app that unlocks distracting apps only after real-world habits — walking, meditating, studying — are completed. Activities are motion-verified and all state stays on the device using the Screen Time / Family Controls framework.
iOS · App Store
Swift · Family Controls · Core Motion · On-device storage
An iPad app for mapping discussion in classrooms, coaching, and meetings. Tap who is speaking and watch turn-taking and timing build up in real time, with session exports and replayable participation insights.
iPad · App Store
Swift · SwiftUI · On-device data · Exports
Turn ambiguous requirements into systems, APIs, data models, and technical boundaries.
Implement backend services, AI workflows, integrations, and product functionality.
Containerize, deploy, automate, and release.
Handle reliability, security, observability, performance, and failures.
Measure real behavior and continuously improve the system.
I care about the part after the demo: making software reliable enough for people to actually use.
LLMs · RAG · Agents · MCP · Human-in-the-loop · Embeddings · Vector Search · Evaluation · Local Inference
Python · FastAPI · Django · PostgreSQL · APIs · Distributed Systems
AWS · GCP · Docker · CI/CD · Observability · Deployment
Swift · SwiftUI · Foundation Models · MLX · Core ML
Authentication · OAuth · Sandboxing · Privacy · Secure APIs
Open-source work, research implementations, and infrastructure projects. Technical projects — not production products.
A fully local retrieval-augmented generation pipeline: ingestion, embeddings, a Qdrant vector store, retrieval, and answer generation with Ollama — no external API calls.
Why it matters: demonstrates an end-to-end RAG system where the data never leaves the host — the pattern behind privacy-first AI products.
A port of MicroGPT from Python to Swift — reimplementing a minimal GPT and its training loop to run as a native CLI.
Why it matters: ML fundamentals implemented from scratch, not just called through an API. Shows understanding of what happens inside a transformer.
A Python module for constraining the execution of untrusted Python code, reducing the blast radius of running user-supplied scripts.
Why it matters: safe code execution is a real systems problem in AI products that run model-generated or user-supplied code.
A multi-stream AirPlay receiver and recorder for macOS: implements the AirPlay protocol to monitor multiple screens at once, with adaptive grid view, MP4 recording, and PIN protection. GPL-licensed.
Why it matters: real protocol-level systems work in Swift, shipped as a usable macOS application.
A Swift package wrapping Uber’s H3 hexagonal hierarchical geospatial indexing library (v4.4.1), exposing a C library cleanly to Swift.
Why it matters: native interop and packaging — bridging a C library into idiomatic Swift for spatial indexing workloads.
A minimal implementation for running Qwen Image generation locally on a Mac — exploring on-device image models on Apple Silicon.
Why it matters: hands-on with local generative models and the realities of running them on consumer hardware.
Selected technical writing on AI systems, local inference, and ML fundamentals.
I’m a software engineer and systems architect with 15+ years of experience building backend systems, cloud infrastructure, and production applications.
My recent work focuses on AI systems: RAG, local LLM inference, AI-powered products, and on-device intelligence. I work across the full engineering lifecycle, from architecture and implementation through deployment and production.
Recent work includes shipping AI-powered iOS applications, building local RAG systems, implementing ML fundamentals in Swift, and developing privacy-first software that keeps data on the device. I’m ISC² certified in cybersecurity, so threat modeling and secure design are part of how I build, not an afterthought.
I care less about using the newest framework and more about whether the system works reliably when real users depend on it.
Open to senior and staff-level engineering roles, consulting, and contract work — remote or hybrid in the EU timezone. If you’re building AI systems that need to run reliably in production, that’s the work I do best.