How we work
Most slow systems are slow by accident. The fix is usually boring.
A missing index, an N+1 loop, a cache nobody invalidates, a spreadsheet someone updates by hand every Monday. We find the boring cause, fix it with the smallest change that works, and prove it with numbers from your own system.
The loop
Baseline. Change. Benchmark.
Every engagement, from a 5-day Diagnostic to a monthly retainer, runs on the same loop.
- 01
Baseline
Before we touch anything, we measure the numbers you care about: p95 latency, query time, cost per request, hours spent on a task.
- 02
Change
One change at a time, as a small PR against your repo, reviewed by your team through your normal process.
- 03
Benchmark
Every change is measured against the baseline. If it didn't move the number, it doesn't ship.
Access & security
NDA first. Read-only by default.
- NDA before access
- We sign your NDA, or send ours, before we see a line of code or a single log.
- Read-only by default
- Metrics, logs, traces, a replica or staging database. We ask for write access only when a specific change needs it, and only for as long as it needs it.
- Nothing in production without you
- Production changes go through your review and your deploy process, on a schedule you agree to.
- Your data stays with you
- We work inside your environment and don't copy production data to our machines. Access is removed when the engagement ends.
Working together
Small team, short feedback loops.
- The people on the call write the code
- Senior engineers only. No hand-offs to someone you haven't met.
- Weekly written update
- What we measured, what shipped, what's next, and anything blocked. Short enough to read in two minutes.
- Async first
- A shared Slack channel or email thread. Calls when they're faster, not by default.
- Everything in your repo
- Code, docs, runbooks and dashboards live in your tools. You own all of it from day one.
On AI
Useful, not magic.
We build AI features and agents for a living, and we'll still tell you when a SQL query, a cron job or a twenty-line script does the job better. When AI does fit, it ships with evals, cost limits and a fallback for when the model gets it wrong.
What we won't do
Things we say no to.
- ×Rewrite your system because we'd have built it differently
- ×Add a framework, vendor or service you'll depend on us to run
- ×Send an invoice you didn't agree to up front
- ×Take a project we don't think we can move the needle on
Start with a 5-day Diagnostic.
$2,500, fixed. Credited toward follow-on work, and your money back if it doesn't find a fix worth the fee.