Skill UI

Configure a Mac mini as a reliable local LLM server with remote access, observability, and power-safe operation. Use when building an always-on private AI inference server on Apple Silicon.

Deploy ML models on Kubernetes with KServe and NVIDIA Triton Inference Server. Covers canary deployments, autoscaling, GPU-aware scheduling, and traffic splitting for production inference workloads.

Generates SQL validation notebooks for dbt PR changes with before/after comparison queries. Supports GitHub PR URLs and local repositories. Creates interactive notebooks for database validation across Snowflake, BigQuery, Redshift, and Athena. Includes schema inference from dbt_project.yml and model configs.

Design secure multi-tenant LLM hosting platforms on shared GPU inference infrastructure, covering tenant isolation, per-tenant quotas and rate limits, billing attribution, noisy-neighbor protection, and per-tenant policy controls for vLLM on Kubernetes.

Unified LLM inference layer built into Neon branches: one credential and one OpenAI-compatible API for frontier and open-source models from seven providers. Switch models without rewiring, wire it into OpenAI/Anthropic SDKs, and keep AI requests branch-scoped.

Guides agents through bounded parameter optimization for configurable systems with measurable objectives: define search space, budget, and baseline, run trials one at a time, log evidence, and recommend a configuration for HPO, inference tuning, simulations, or RL/control experiments.

This skill manages the entire lifecycle of long-running deep learning training and inference jobs deployed on remote, ephemeral computing environments (e.g., AutoDL, RunPod, bare SSH, Slurm, K8s). It provides critical capabilities like resilient checkpointing, managing spot instance preemption, debugging resource failures (OOM, NaN), and ensuring safe billing teardown, making it essential for professional ML Ops on rented resources.

A critical discipline for managing expenses when integrating paid AI or inference APIs. This guard treats dollar cost as a third complexity dimension, mandating the definition of cost contracts (max calls per run, max budget per run, max budget per day). It enforces best practices for bounding loops, retries, and fan-out, ensuring that code bugs cannot lead to unexpected overspending. Best practice requires configuring hard caps both in the code and at the provider dashboard for defense in depth.

Deploy and manage vLLM for high-throughput LLM inference. Configure continuous batching, tensor parallelism, quantization, and OpenAI-compatible API endpoints for production LLM serving.

Master the Zod library for defining robust, type-safe schemas in TypeScript. This expert covers advanced validation techniques, including primitives, objects, unions, and discriminated unions. It specializes in type inference (`z.infer`), data coercion, complex refinements, and seamless integration with modern frameworks like React Hook Form, Next.js, and tRPC, ensuring consistency across your application state and API inputs.