Written like incident reports, because that's what we trust

situation

Their order system ran on a single database server — the classic setup. It had been "fine for years," which is what every single point of failure says right up until it isn't.

what we built

• Replaced the lone server with a clustered setup: if one machine fails, a standby takes over automatically
• Spread the cluster across two locations so even a site problem doesn't stop orders
• Added monitoring that pages our on-call engineer the moment anything looks off
• Rehearsed the failure on purpose, repeatedly, until the recovery was boring

result

Months later, the main server died at 2:14am — real hardware failure, not a drill. The standby took over in 4.2 seconds. No orders lost, no late-night calls, no morning crisis. The client read about it in our incident note over coffee.

situation

A school district's worst IT day is always the same one: the first day of school. Thousands of students and staff log in at once, every classroom needs wi-fi, and any system that limped through summer fails at 8:05am in front of everyone. There are no do-overs on a start date.

what we built

• A summer readiness program: every classroom device refreshed, tested, and ready before staff returned
• Wireless coverage reviewed and reinforced at the campuses that struggled the year before
• Monitoring on the systems that matter at 8am — sign-ins, wi-fi, classroom tools — watched live during opening week
• A surge plan for our help desk so the first-week flood of tickets got answered in minutes, not days

result

School opened on schedule with every classroom logging in normally. The first-week ticket queue stayed under control, response times held, and the district's staff spent opening week teaching instead of waiting on hold.

situation

The practice ran on its practice-management software, hosted on an aging server in a closet. Backups existed in theory, nobody had ever tested a restore, and the HIPAA security paperwork was a folder of good intentions. Their cyber-insurance renewal questionnaire is what finally forced the question.

what we built

• Encrypted, off-site backups with scheduled restore tests — and reports the practice can hand to an auditor
• Multi-factor authentication on email and every administrative login
• Same-day access shutoff procedures for staff departures, documented
• A written security plan mapped to HIPAA requirements, with evidence filed for each control, under a signed business associate agreement

result

The insurance questionnaire went from a fire drill to an afternoon. Then we ran a full recovery drill: simulated server loss, restore from backup, practice back to scheduling patients in 45 minutes. The owner watched it happen — which is the point.

situation

Their ERP — the system that takes every order — was a heavily tailored Odoo setup, several versions behind. The previous vendor quoted a multi-week freeze on changes and couldn't say what would break. For a plant that enters orders every business day, that answer was the problem.

what we built

• Rebuilt the upgrade on a copy of their system first, fixing everything that broke where it couldn't hurt anyone
• Ran two timed full-dress rehearsals with their key people clicking through real daily workflows
• Scheduled the real cutover for a weekend: freeze Friday evening, upgrade Saturday, their team verifying Sunday
• Kept a back-out plan armed through the following week, just in case

result

Cutover confirmed Sunday at 4pm, two hours ahead of the deadline. Monday at 7am the order desk opened on the new version. No order-entry interruption, no emergency fixes in week one, and the back-out plan was retired unused.

situation

Vendor invoices, maintenance requests, and tenant emails all arrived in shared inboxes. Two employees spent a combined ~40 hours per month opening attachments, retyping numbers into the accounting system, and forwarding mail to whoever seemed right. Documents went missing; data entry lagged reality by days.

what we built

• An automated pipeline that watches the shared inboxes and reads each document as it arrives
• Invoices matched to the right property and vendor, entered as drafts a person approves
• Maintenance requests turned into work orders in the right queue, with due dates
• Anything the software isn't confident about flagged for human review — not guessed

result

Roughly 40 hours a month of manual entry eliminated. The books now lag reality by hours instead of days, and the two employees moved to work that needs judgment. About one document in twenty still gets human eyes — exactly the ones that should.