All of it on one page. Jump to the area you care about, then use the links on each card for the full story.
The hands-on side — diagnosing and rebuilding real physical systems, from the wiring in the wall to the app that runs on top of it.
Replacing a failed intercom in a historic limestone apartment building, plus one web app with three sides: the public site, residents, and staff. The door work is paused by the client’s budget until 2027; the website build continues. Read the case study.
Designing small-business and residential networks and then installing them — cabling, terminations, right-sized IP plans, and mapping undocumented systems with a meter instead of a guess.
The systems everything else runs on — designed, deployed and operated end to end.
Managed virtual private servers joined to a private mesh network, with automated provisioning, identity integration, monitoring, backup and a documented exit path. Customers get their own environment without operating it themselves. Read the case study.
A control plane that takes a new node from bare host to a fully configured, TLS-terminated, SSO-integrated application stack without manual steps — including per-job tokens, health verification and rollback.
Centralized authentication across every hosted application using Authentik, with OIDC clients, federated login and per-service access policy.
Self-operated detection and vulnerability management — SIEM, host monitoring, intrusion prevention and scheduled scanning across the estate.
Off-grid and resilient communications devices.
Off-grid communication hardware for Meshtastic and MeshCore networks — solar nodes, repeaters and portable kits built for operation with no infrastructure, no subscription and no cell coverage.
A single mesh device combining LoRa messaging, local services and a self-contained power system — intended for deployment where conventional connectivity is unavailable or unreliable.
Applied AI systems, built and run on owned hardware rather than rented APIs.
An agent platform and orchestration backend running against local models on owned GPUs, with a tool-calling runtime, job scheduling and a web interface.
Retrieval-augmented knowledge infrastructure — a vector-backed document store and API used to ground AI systems in real organizational content instead of guesswork.
A queueing layer that shares a finite pool of GPUs across inference, image generation and training workloads without jobs trampling each other.
A real-time news monitoring platform. It collects articles and posts, streams them through Kafka and Apache Flink, and uses local AI models to tell genuine trending stories apart from coordinated or misleading content. The heavy processing runs overnight on systemd timers, and the results feed the content pipelines below.
An automated social-media pipeline. It picks a timely topic — often a current story from Falken — writes the caption in each brand’s own voice, generates the image on a local GPU, and posts to Mastodon and Bluesky on a natural, human-paced schedule. Several brand accounts stay active every day with nobody at the keyboard.
A daily world-news podcast made start to finish by an automated pipeline. Every morning it ranks the day’s stories from the 700+ news sites Falken follows, GDELT’s global news coverage and social media by how widely each one is spreading outside the US, writes the script with a local language model, voices it and publishes it — about ten minutes, no hands. A watchdog checks every run and alerts or repairs on a failure.
The video side of the same pipeline. Each episode automatically becomes a video — an animated on-screen host set in a retro TV frame, rendered through the GPU Scheduler — then uploads to a self-hosted PeerTube channel and gets promoted by the Meme Engine. One script in; a podcast, a video and social posts out.
Production applications built for real customers and real transactions.
A full e-commerce storefront on a headless commerce backend, with print-on-demand fulfillment integration, custom React front end and automated order sync.
A federated social platform with on-device translation, multi-protocol follow adapters and a video pipeline — built to interoperate with existing networks rather than replace them.
An artist platform with provenance tracking, verification, commissions and federation services running as independent microservices.
A local barter and marketplace where the unit of exchange is not always money. People list what they have and what they want, and the system matches trades within a real geographic area, favoring nearby exchange over shipping across the country.
Self-hosted Mastodon, Pixelfed, PeerTube, Matrix and Bluesky PDS instances — operated as a single interconnected estate with shared identity and monitoring.
Communication for an audience that has no patience for jargon.
A published children’s picture book — a true story about a three-legged deer, written under the pen name Jeffrey Lebowski. Different audience, same discipline: take something real and make it land for the reader in front of you.
A short, playable version of one of the games I use with teams. Nothing you do is recorded.
The method behind every engagement, and the hosting side of the practice.
Observe, measure, map, identify, experiment, measure again. Recommendations get tested rather than defended.
Open How It Works →Find out how much the people in your company disagree when they look at the same thing, what kind of disagreement it is, and what to do about it. You may not have a people problem. You may have a noise problem.
Managed private servers on a mesh network. The hosting product has its own site, because it is a product and this is a consulting practice.
Open Infrastructure →Things that come up often enough in engagements to be worth writing down properly.
Notes on systems, noise and decisions. Every article, in one list.
Open Thinking →Before you cancel the recurring meeting, find out what it was quietly doing.
Open Meetings Are Often a Symptom, Not the Problem →How callers described a problem told me exactly where they were standing.
Open What Technical Support Taught Me About Organizational Design →The hardest recommendation to charge for is the one where nothing gets purchased.
Open Sometimes the Correct Recommendation Is: Don’t Buy Anything →Tell me what is not working. I will tell you which part of this is relevant, if any of it is.
