Field notes from Pubful: practical engineering on AI systems, infrastructure, HPC and GPU clusters, software, and hardware.
-
Build, rent, or burst: right-sizing GPU capacity for AI workloads
A practical framework for deciding when to own GPUs, when to rent them, and when to burst to the cloud — without overspending or stalling your roadmap.
Read → -
Bare metal at scale: provisioning and operating servers across many sites
Running bare-metal fleets across multiple locations is a provisioning and operations problem, not a hardware one. How to make racks of servers behave like one programmable system — from network boot to multi-site operations.
Read → -
From prototype to production: shipping an AI feature that holds up
A working demo is the easy 20%. Here is the engineering that turns an impressive prototype into an AI feature you can put in front of real users — and keep running.
Read → -
On-prem, colocation, or cloud: where your infrastructure should actually live
Cloud-by-default is a decision, not a law of nature. A clear framework for choosing on-premises, colocation, or cloud — and why the answer is usually a deliberate mix.
Read → -
Where cloud bills go wrong — and how to cut them without breaking things
Most cloud overspend is structural, not wasteful clicks. A FinOps and SRE view of the real cost drivers and how to cut them safely while keeping reliability intact.
Read → -
Beyond IoT: the platform layer that runs a fleet of bespoke devices
Connecting one device is a weekend project. Running thousands securely, updating them safely, and onboarding new ones with zero touch is a platform problem. Here is what that platform needs.
Read → -
Connecting a modern vessel: NMEA 2000, Signal K, and remote fleet monitoring
A practical guide to marine systems integration — how NMEA 2000, NMEA 0183, J1939, and Signal K fit together, and how to get vessel data ashore for remote monitoring.
Read → -
Making medical imaging systems talk: DICOM, HL7/FHIR, and HIPAA-safe pipelines
How modalities, PACS, and clinical systems actually interoperate — DICOM and DICOMweb, HL7 and FHIR, IHE workflows, de-identification, and building it HIPAA-consciously.
Read → -
What a software-defined vehicle actually requires: networks, middleware, and OTA
The software-defined vehicle in concrete terms — in-vehicle networks (CAN FD, automotive Ethernet), service middleware (SOME/IP, DDS, AUTOSAR), zonal architecture, OTA, and UDS/DoIP diagnostics.
Read →