Work that has shipped, scaled, and held up under production load.

OpenClaw Cloud is a multi-tenant SaaS platform that wraps the open-source OpenClaw agent engine and delivers a personal AI assistant across the chat applications people already use — WhatsApp, Telegram, Slack, and Discord. It is live in production and serves thousands of users.

The system is built as a Turborepo monorepo with four applications: a Next.js 15 dashboard (Vercel), an Express + WebSocket gateway that runs the OpenClaw engine (Fly.io), a webhook receiver, and a background worker. A dedicated cloud infrastructure layer handles authentication, Stripe-based billing and credits, team management, multi-tenant isolation, RBAC, SSO/SAML, and AES-256-GCM key encryption.

The platform exposes a public REST API, cron scheduling, custom domains, and a plugin system, while the core engine — consumed as a workspace dependency rather than forked — provides 55+ skills, session management, multi-model support with fallback chains, function calling, memory, and code execution.

LokApp is a published consumer mobile application available on the Apple App Store and Google Play. It lets users catalog physical belongings by photographing them — AI identifies each item, extracts metadata (name, category, brand, model, material), and stores it in a searchable organized inventory.

Users can scan entire shelves or boxes in a single photo, ask natural-language questions such as "where is my blue winter jacket?", generate QR code labels for containers, track item movement history, and share locations with up to six members across owner/contributor/viewer roles.

Built with React Native on Expo SDK 55, the app uses Supabase for data and authentication, Cloudflare R2 for compressed photo storage and CDN delivery, and Postgres full-text search with tsvector. Tiered subscriptions and a no-expiry AI credit top-up round out the product.

Domu.care is a cross-border digital care platform that connects experienced caregivers from Croatia, Serbia, and Slovakia with families in Austria who need 24-hour home care for elderly relatives. It operates purely as a software provider — families and caregivers contract directly, while the platform handles matching, scheduling, onboarding paperwork, and coordination.

Families build household and resident profiles, get matched with qualified caregivers by language and experience, manage 2–4 week rotation cycles with handover continuity, and track medication and daily journals. Caregivers set their own daily rate, keep 100% of their earnings, and coordinate travel through shared transport groups.

The platform uses LLM-assisted conversational onboarding (AI assistants named Alma for families and Mila for caregivers), a versioned subscription pricing engine with admin-controlled discounts, an action-logged admin console, and a per-user token usage dashboard spanning multiple LLM providers.

Patrona is a Fractional CTO engagement covering the end-to-end architecture of a platform that supports victims of domestic abuse. Because of the sensitivity of the users served and the evidentiary nature of the data collected, the project operates under NDA — architectural details, product features, and operational specifics are intentionally withheld here.

What can be shared publicly: the system is a cross-platform solution spanning cloud infrastructure, a native iOS companion, and a web console. Stack highlights include AWS (CDK, Lambda, S3, DynamoDB, Cognito, Secrets Manager), Swift 6 with strict concurrency on iOS, a React web SPA, and client-side encryption across every tier so plaintext never reaches the server.

The engagement covers technology strategy, architecture, security posture, compliance, cloud operations, and hands-on implementation — representative of the deep technical leadership provided under a fractional CTO agreement.

ScanDoc is a bilingual (Italian / English) document intelligence SaaS built for the Italian market. It ingests unstructured scanner output from NAS or SFTP endpoints, classifies business documents — fatture, lettere, offerte — using vision models, extracts structured metadata, and auto-files everything into a clean Customer / Year / Type folder hierarchy.

The extraction pipeline validates Italian tax identifiers (Partita IVA check digit, Codice Fiscale) and performs a three-step customer matching cascade: exact Partita IVA match, pg_trgm fuzzy name similarity, and email domain match. This matching layer feeds directly into the customer's CRM system to link every filed document to the right account.

The architecture is FastAPI + Celery workers with Supabase Postgres, AWS S3 and SQS for event-driven processing, OpenRouter vision models for classification, and Upstash Redis. Low-confidence documents route to an operator review queue whose corrections are fed back as few-shot examples, improving recognition accuracy over time.

A professional vehicle diagnostic system for FCA / Stellantis platforms — deep enough to perform full ECU scanning, fault-code reading and clearing, live data monitoring, security-access seed capture, and brute-force key discovery on specific control units.

The architecture separates hardware and UI across two applications. A Windows desktop built on Python and PySide6 handles the full UDS / OBD protocol stack, service workflows, and dashboard UI. A companion Android agent, written in Kotlin with Jetpack Compose, bridges Bluetooth diagnostic adapters (OBDLink MX+, ELM327, and similar) and proxies commands to the desktop over a JSON-RPC 2.0 channel on WebSocket.

The two sides can communicate on the local network or across the internet via a zrok tunnel with QR code pairing — enabling remote support sessions — and the desktop ships with a full simulation mode for testing workflows without hardware present.

A custom IoT firmware project that delivers branded touchscreen controllers for Home Assistant. The hardware target is a 320×480 QSPI TFT display driven by an ESP32-S3 with 16 MB flash and octal PSRAM, paired with an AXS15231 capacitive touchscreen.

The interface is an LVGL-based seven-page layout: a Home dashboard of quick-access buttons, a climate page with per-room temperature and humidity, an energy page with real-time per-circuit power consumption, a thermostat page, two lighting control pages, and a door / gate / automation page.

User experience details include swipe navigation between all pages, wake on touch or motion, auto-sleep after 20 seconds with burn-in prevention, and live sensor updates driven directly from Home Assistant over its API — a practical example of building production-grade embedded interfaces around an existing home automation platform.