Expert in Machine Learning Operations bridging data science and DevOps. Use when building ML pipelines, model versioning, feature stores, or production ML serving.
Fine-tune LLMs and ML models — LoRA, QLoRA, PEFT, Hugging Face. Dataset prep, training, evaluation, deployment
Model Registry Manager - Auto-activating skill for ML Deployment. Triggers on: model registry manager, model registry manager Part of the ML Deployment skill category.
Generate a reproducible, runnable PyTorch training repo for a medical-imaging task — segmentation, classification, detection, image-to-image synthesis, self-supervised…
Deploy ML models as production APIs with vLLM, TGI, ONNX Runtime, batching, autoscaling, and GPU optimization.
Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense models), imple — from…
Use when debugging a Nemo Gym run or reward profiling job. Covers rollout collection failures, empty or partial JSONL outputs, stale materialized inputs, verifier/schema errors,…
Use when building GraphRAG pipelines on Neo4j with the neo4j-graphrag Python package: VectorRetriever, VectorCypherRetriever, HybridRetriever, HybridCypherRetriever,…
Use when you have raw mzML files and feature tables (CSV from mzMine or XCMS) for LCMS data, have generated training/validation/test batches with known class imbalance, and need…
Use when you have a working base MPNN model (e.g., chemprop) and need to add task-specific feature processing layers (spectral, electronic, or domain features) to improve…
Use when you have a pretrained TCN spectrum encoder from formula prediction and need to train a rescoring model that ranks formula candidates by confidence.
Use when after implementing a neural network component that will feed into a downstream architecture (e.g., a transformer).
Activates the 'Swarm Orchestrator' for building multi-agent systems with the openai-agents SDK. Enforces a 'Live Docs First' approach to verify syntax via Context7, preventing API…
Use when implementing retrieval-augmented generation with OpenAI from an approved ai-architecture.md and retrieval design.
Use when implementing an OpenAI-backed AI capability that must return schema-bound JSON, typed objects, classifications, extraction results, or other machine-consumable responses…
Use when implementing OpenAI tool or function calling from an approved ai-architecture.md tool surface.
Fine-tune and serve Physical Intelligence OpenPI models (pi0, pi0-fast, pi0.5) using JAX or PyTorch backends for robot policy inference across ALOHA, DROID, and LIBERO…
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3.
Fine-tunes and evaluates OpenVLA-OFT and OpenVLA-OFT+ policies for robot action generation with continuous action heads, LoRA adaptation, and FiLM conditioning on LIBERO…
Orchestrate end-to-end machine learning pipelines using Prefect or Airflow with DAG construction, task dependencies, retry logic, scheduling, monitoring, and integration with…
Integrates with the Plaid Transactions API using the plaid Python SDK to pull 90 days of transaction history across linked bank accounts.
Query, discover, and enrich the PMOVES.AI model catalog. Manages all AI model metadata (LLM, embedding, TTS, vision), HuggingFace enrichment, TensorZero TOML config export, and…
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3.
Align a fine-tuned model with preference data using DPO, ORPO, KTO, or SimPO. Use when preference pairs or thumbs-up/down feedback exist, when choosing between…
Guarantee valid JSON/XML/code structure during generation, use Pydantic models for type-safe outputs, support local models (Transformers, vLLM), and maximize inference sp — from…
Expert guidance for PyTorch Dynamo (torch.compile) development and debugging. Covers symbolic execution, VariableTracker system, pytree integration, guard failures, guard…
Use when after defining a transformer encoder architecture with multi-head self-attention and positional encoding, and before training or inference on mass spectrometry data.
中文优先：用于PyTorch模式相关任务，帮助识别、设计、实现或验证对应工作流。English keywords: PyTorch deep learning patterns and best practices for building robust, efficient, and reproducible training pipelines,…
Optimize accuracy for RAG (Retrieval-Augmented Generation) systems. Covers: DB schema design, chunking strategies, retrieval optimization, accuracy testing, and anti-hallucination…
Design LLM applications using the LangChain framework with agents, memory, and tool integration patterns.
Review an existing Retrieval-Augmented Generation system and find why it underperforms. Use when asked to review or audit a RAG pipeline, diagnose wrong/ungrounded answers from a…
Evaluates RAG pipeline quality across retrieval (precision, recall, MRR) and generation (groundedness, hallucination rate).
Build production document ingestion pipelines with chunking, embedding, and vector DB storage. Activate on: document ingestion, chunking strategy, embedding pipeline, vector DB…
Retrieval-Augmented Generation patterns on Oracle Cloud Infrastructure — embeddings, vector stores, hybrid search, reranking, and production RAG architecture
RAG architecture: embeddings, chunking strategies, hybrid search (BM25 + vector), reranking, CRAG/self-correcting, multi-hop reasoning, evaluation metrics.
Design and architect RAG (Retrieval-Augmented Generation) pipelines. Covers vector DB selection, chunking strategies, hybrid retrieval (vector + knowledge graph), semantic…
High-performance vector similarity search engine for RAG and semantic search. Use when building production RAG systems requiring fast nearest neighbor search, hybrid sear — from…
Domain routing and boundary guidance for authorized clickjacking testing, including missing X-Frame-Options, CSP frame-ancestors bypasses, and drag-and-drop hijacking.
Run repeatable eval suites against prompts, RAG pipelines, and agents so regressions surface before release.
Use promptfoo when an agent needs to evaluate prompt, agent, or RAG behavior against saved assertions before a change goes live.
ReWOO decoupled planning — Planner/Worker/Solver split. 5x fewer tokens than ReAct on HotpotQA, +4% accuracy. Plan-and-Execute generalization, planner distillation to 7B.
Curriculum AI engineering từ toán học lên production — 503 bài học, 20 phase, 320 giờ. MIT license. Build từ first principles: backprop, transformer, LLM, agent, swarm.
Self-Refine iterative improvement (generate/feedback/refine loop, +20 avg across 7 tasks) and CRITIC external verification (tool-grounded critique for factual tasks).
World-class computer vision skill for image/video processing, object detection, segmentation, and visual AI systems.
World-class computer vision skill for image/video processing, object detection, segmentation, and visual AI systems.
World-class ML engineering skill for productionizing ML models, MLOps, and building scalable ML systems.
World-class ML engineering skill for productionizing ML models, MLOps, and building scalable ML systems.
Fast structured generation and serving for LLMs with RadixAttention prefix caching. Use for JSON/regex outputs, constrained decoding, agentic workflows with tool calls, o — from…
sGLang RadixAttention — KV cache stored in a radix tree, reused across requests sharing common prefixes. Cache-aware scheduling (depth-first, LRU at branch level).
Skill library patterns from Voyager — executable code as skills, semantic retrieval, composition, failure-driven refinement.
Build lightweight AI agents with HuggingFace Smolagents — use CodeAgent (writes Python to act) or ToolCallingAgent (JSON tool calls), add built-in or custom Tools, orchestrate…
Integrate a HuggingFace Computer Vision model into the NVIDIA TAO Toolkit ecosystem (tao-core config, tao-pytorch trainer, tao-deploy TensorRT pipeline).
Tensorflow Savedmodel Creator - Auto-activating skill for ML Deployment. Triggers on: tensorflow savedmodel creator, tensorflow savedmodel creator Part of the ML Deployment skill…
Tensorflow Serving Setup - Auto-activating skill for ML Deployment. Triggers on: tensorflow serving setup, tensorflow serving setup Part of the ML Deployment skill category.
Probe RAG applications for prompt injection via poisoned retrieved context and embedding manipulation.
Install Together AI SDK and configure API key for inference and fine-tuning. Use when setting up Together AI, configuring the OpenAI-compatible API, or initializing the together…
Instrument Python or TypeScript agent applications with OpenTelemetry traces, metrics, and evaluation views for debugging production LLM workflows.
Semantic codebase and architecture-decision discovery with vaultspec-rag - find code and the ADRs that govern it by meaning, then narrow with advanced filters and noise controls.
Migrate vector data between Pinecone, Qdrant, Weaviate, pgvector with re-embedding and schema mapping.
Server-side extension that completes the full analysis pipeline for image classification after vera-ai-image-testing has run.