AI Projects › Technical skills
This is the hands-on side. Every item is something I have built, run or repaired, with the page that shows the evidence.
Storage and servers
- TrueNAS SCALE and ZFS across multiple pools, with a single-disk-pool risk map.
- Rebuilding a failed five-drive NAS by hand and checking every share against its “already migrated” copy.
- Diagnosing a boot-loop crash from console photos, cancelling an unnecessary reformat and salvaging the unique data.
- Evacuating a failing drive (118,599 files, 1.52 TB) with every count, byte and date verified.
- Working with awkward older gear, including a forgotten NAS with one weak core. See hardware limits.
Networking and security
- UniFi networks, cameras and door access.
- One front door for public access: a single forwarded port pair into a hostname-routed reverse proxy, instead of per-server forwards.
- Read-only mounts, least privilege and logged approvals.
- Finding the real bottleneck by measurement: a 1 gigabit mini-PC throttled a 10 gigabit server to about 8% of its link.
Data and databases
- A Postgres inventory of every file: sha256, md5, quick hash and a media probe, written in real time.
- Three quality gates per volume: counts match, every file hashed, a random independent re-hash agrees.
- Deduplication, golden-record selection and parent-to-child lineage.
- pgvector groundwork for search by meaning. See the six AI skill areas.
Video and media
- A decade or more of video and drone production.
- GPU HEVC encoding with VMAF quality scoring and triage rules.
- Metadata care: 360 degree flags, GPS and telemetry tracks, spatial audio and original timestamps. See the two nights.
- Premiere and Media Encoder workflows.
AI operations
- A manager agent and two worker agents, with job cards and check-backs.
- A five-level risk ladder that decides what AI does alone and what a human must run. See guardrails.
- Orchestration in plain scripts that use no model tokens.
- Knowing when not to train a model on messy data.
Marketing and sales engineering
- Decades of positioning, messaging and customer storytelling.
- Demos, events and testimonials, and the data behind them.
- Turning technical proof into something a buyer can act on.