Production profiling engagement
On-call week with your repo, representative traffic or fixtures, and a written hotspot report with patch proposals.
Engagement outline →When your Python service is fast in staging and stubborn in production
We run structured profiling sessions on your real workloads, read the flame graphs with your engineers, and land focused patches—usually within a two-week engagement window.
What teams book us for
Most calls start with a latency spike, a worker that grows in RAM overnight, or a batch job that missed its window. We narrow the question before touching code.
Hotspot code review
Half-day review of modules you already suspect—ORM loops, serialization, or asyncio gather patterns.
Review scope →
Load test readout
We join your k6 or Locust run and map failures to Python call stacks instead of guessing from dashboards alone.
What we need from you →A recent Cape Town fintech handoff
Their Celery queue backed up every market open. We profiled one worker under recorded tasks, found a pandas concat inside a per-message loop, and replaced it with pre-allocated buffers. Queue depth returned to normal without adding hardware.
Engagement length
Nine business days including two on-site afternoons at Adderley Street.
Deliverables
Annotated cProfile exports, three merged PRs, runbook for weekly regression checks.
Follow-up
Optional office hour thirty days later to re-profile after their peak season.
“The report named the exact SQLAlchemy loader we argued about for weeks. We still had to schedule the migration ourselves, but at least we stopped debating whether the problem was Postgres or Python.” — Lindiwe N., backend lead, logistics API (Johannesburg)
From the Field Notes
Short guides we send to clients before the first profiling session.