Marketplace

Design recommendation systems: collaborative filtering, content-based, hybrid approaches, cold start, and ranking.

Build retrieval-augmented generation pipelines: chunking strategies, embedding models, vector stores, and reranking for accurate grounding.

Establish versioned prompt management: semantic versioning, A/B testing, rollback, and changelog practices for production prompt systems.

Design comprehensive observability: metrics, logs, traces, and profiles with SLO-driven alerting and dashboard design.

Design multi-region active-active and active-passive architectures: data replication, failover, DNS routing, and consistency.

Build comprehensive model evaluation: benchmark selection, statistical significance, human evaluation protocols, and safety testing.

Plan ML model deployment: serving patterns, A/B testing, canary releases, monitoring, and rollback for production ML.

Design zero trust network architectures: identity-based perimeters, microsegmentation, and continuous verification.

Design and implement ML feature stores: online/offline serving, feature freshness, training-serving skew prevention, and feature discovery.

Design ML experiment tracking: hyperparameter logging, model versioning, reproducibility, and experiment comparison workflows.

Design disaster recovery strategies: RTO/RPO targets, backup automation, failover procedures, and DR testing.

Optimize how context is assembled for LLM calls: system prompt design, few-shot example selection, context pruning, and token budget management.

Scan and harden container images: base image vulnerabilities, runtime permissions, distroless patterns, and admission policies.

Audit application configuration: environment variables, secrets management, config validation, and environment parity.

Audit code for race conditions, deadlocks, and thread safety issues in async and multi-threaded code.

Harden APIs against attacks: input validation, output encoding, rate limiting, CORS, CSP, and injection prevention.

Build automated test suites that exercise agent behavior across edge cases: adversarial inputs, tool failures, multi-turn conversations, and recovery.

Analyze and reduce AI agent operating costs: model routing, prompt caching, context window management, and batch processing strategies.

Use when building or auditing an LLM-as-judge: calibrate it against human labels, measure agreement, and de-bias position and verbosity effects.

Invoke to audit or tighten an AI agent's tool/MCP permissions: diff granted access against observed usage and recommend a least-privilege allowlist.

Invoke when an AI system touches the EU market and you must classify its EU AI Act risk tier and output the obligations and article citations that follow.

Use when a repo's test suite is red and you must drive it to green by patching the code under test within a budget, without masking regressions.

Design secure execution environments for AI agents: Docker sandboxes, API key isolation, network policies, and resource limits.

Craft detailed agent personas with expertise areas, communication styles, decision frameworks, and behavioral boundaries.