Python placements cover wildly different work shapes — Django web backends, FastAPI service tiers, ML pipelines (NumPy/Pandas/PyTorch/scikit-learn), data engineering (Airflow/dbt/Spark), automation scripting. sourceBOLD’s 4-gate technical screen calibrates per-stack: the rubric for a backend Django role differs from the rubric for an ML-pipeline role. Same gauntlet, different question batteries.
60-minute structured exercise calibrated against the target stack. A Django/FastAPI backend candidate gets exercises on request handling, ORM patterns, serialization, async vs. sync boundaries. A data-engineering candidate gets exercises on schema design, throughput, idempotency. Two reviewers, blind to candidate identity, must agree to pass.
Architecture round, also stack-dependent. Backend: partitioning, caching, idempotency, async patterns. Data engineering: throughput, fault tolerance, schema evolution, late-arriving data. ML: training pipelines, evaluation discipline, drift detection. Same set of questions for each stack; calibrated against engineers already placed.
Two-hour real-codebase session against a benchmark repo in the candidate’s stack. We watch how they handle a failing test, how they reach for documentation, how they reason about edge cases under load. Stack-specific judgment is what G4 catches that résumés don’t.
Full funnel walkthrough at /guides/4-gate-vetting. The per-stack rubric calibration is what makes the 4.2% acceptance rate meaningful across different Python specializations.
The contracting layer sits underneath all of this. From your team’s day-to-day, the developer is a senior Python contractor in Slack and on PRs. The cross-border tax shape, the bi-weekly invoice cycle, the Wise dispatch in the developer’s local currency — those are sourceBOLD’s operational concerns, not your team’s.
A succinct scoping call. Pre-vetted candidates. First standup in a few short weeks.