编程开发

End-to-end latency optimization for an API using flamegraphs + query analysis + caching; includes capacity math.

Correct Redis usage: cache patterns, distributed locks done properly, rate limiters, and failure behavior.

Designs access-pattern-first single-table/denormalized models — the opposite habit from relational design.

Chooses correct locking/transaction patterns for specific race conditions; maps anomalies to concrete business bugs.

Operational Postgres expertise: health diagnostics, indexing, vacuum/ bloat, connections, and replication.

Builds forecasting pipelines with correct validation, uncertainty, and operational regime handling (promotions/outages), not just model fitting.

Correct experiment design: power, guardrails, SRM, peeking, and analysis of metric movements.

Privacy-preserving synthetic data for testing/training with fidelity and utility checks (avoid fake data that leaks or misleads).

Productionizes prompt changes: versioned prompts, offline eval gates, online shadows, and rollback — prompt-as-code.

Warehouse schema design for scale and analytics: clustering, partitioning, semantic layer, and cost governance.

Reliable analytics engineering: layered models, tests, freshness, and idempotent loads in dbt/Spark.

Production ML monitoring: data/concept drift, staleness, fairness over time, and retraining triggers — with alert thresholds tied to impact.

Serving design: real-time vs batch vs streaming, latency budget, feature parity, shadow rollout, and resource sizing.

Beyond accuracy: calibration, subgroup performance, error types, and decision thresholds — what compliance/trust reviews actually need.

Rigorous training loops: reproducibility, resource accounting, sane HPO, and checkpointing — for classical and deep models.

Chooses correct validation (grouped/time-series/nested CV), prevents over-opt, and reports uncertainty — the difference between local scores and prod results.

Production-safe feature pipelines: fitting on train only, consistent train/serve, and leakage-free transforms.

Structured EDA with statistical thinking — distributions, leakage, correlations, quality issues — ends with modeling implications, not pretty-but-empty charts.

Before modeling: defines the business problem, label, metrics, baselines, and data requirements — prevents solving the wrong problem with a fancy model.

Cold-start-aware, stateless serverless design: idempotent functions, fan-out, local emulation, and cost controls.

Secure desktop wrappers: context isolation, IPC validation, auto-update, and installer/code-signing.

MV3 extension architecture: service worker lifecycle, content scripts, storage, CSP restrictions, and store review gotchas.

Cross-platform app with native bridge, push, updates, and platform-specific parity.

Secure, reliable inbound webhook endpoints and outbound webhook delivery system.