Case study / Client engagement, anonymized
Found the network faults a dedicated team could not
A newsroom office suffered intermittent Wi-Fi failures that resisted diagnosis. The dedicated network team had not isolated a cause. I built the telemetry that found it.
Deployed to the fleet- Fleet-wide via MDM
- Zero server infra
- 3 faults isolated
The engagement
Challenge
Intermittent failures, no evidence, a team stuck
The Wi-Fi failed intermittently and the failures resisted diagnosis. A dedicated network team had not isolated the cause, because nobody had per-device, per-access-point evidence of what was actually happening at the moment things broke.
Action
Instrument the fleet, then let AI read the logs
I built a daemon-triggered Mac telemetry system with Claude Code and deployed it fleet-wide via MDM, capturing per-device, per-access-point diagnostics with zero server infrastructure using a resumable uploader. Then I ran AI log analysis across the captured data.
Result
Three overlapping faults, from evidence
The log analysis isolated three overlapping network faults the dedicated network team had not found. Not intuition, not trial and error: the instrumentation produced the evidence and the analysis found the pattern in it.
How it is built
Evidence over intuition
No server was stood up. The uploader streams to a form-backed sheet, so the whole system is zero infrastructure.
- 01Trigger. A LaunchDaemon captures diagnostics on the events that matter, not on a fixed poll that misses the failure.
- 02Capture. Per-device, per-access-point signal and connection diagnostics, the granularity the network team lacked.
- 03Upload. A resumable uploader with no server behind it, so a laptop that drops offline mid-capture does not lose the record.
- 04Deployment. Pushed to the whole fleet through MDM, silently, no per-machine setup.
- 05Analysis. AI correlation across the fleet's logs surfaced three faults that only showed up when the evidence was seen together.
What it proves
The right instrumentation plus AI analysis beat a dedicated team's intuition, because evidence beats guesswork.
See the rest of the work →Client and infrastructure details are anonymized. No client name, network identifiers, or endpoints are shown.