Machine Learning Skills

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

CPU/Memory profiling, bottleneck analizi, benchmark teknikleri ve performans optimizasyonu rehberi.

Statistical modeling toolkit. OLS, GLM, logistic, ARIMA, time series, hypothesis tests, diagnostics, AIC/BIC, for rigorous statistical inference and econometric analysis.

Design LLM applications using the LangChain framework with agents, memory, and tool integration patterns. Use when building LangChain applications, implementing AI agents, or creating complex LLM workflows.

Build Retrieval-Augmented Generation (RAG) systems for LLM applications with vector databases and semantic search. Use when implementing knowledge-grounded AI, building document Q&A systems, or integrating LLMs with external knowledge bases.

Build HTML data explorers for CSV, JSON, log, and event data — filterable tables, faceted search, inline charts, timeline scrubbing, A/B test result dashboards. Use whenever the user has a dataset to explore, browse, filter, chart, facet, or analyze — especially for ad-hoc analysis where opening a heavy BI tool is overkill. Reach for this when the user pastes data, mentions a CSV/JSON/log file, or asks to "look at" or "analyze" a dataset. Always runs a secret-redaction pass before embedding data — credential-shaped values (API keys, tokens, cookies, passwords) are replaced with placeholders unless the user explicitly opts in to embedding them.

Create HTML playgrounds with sliders, knobs, toggles, and live preview for tuning algorithm parameters, animation values, design tokens, layout dimensions, or any value that's painful to express in text. Always include a Submit button (calls `submitToClaude`) so chosen values can be sent back to Claude Code. Use whenever the user wants to experiment with values, fine-tune behaviors, explore a parameter space, or pick from a continuous range — debounce timings, color values, easing curves, threshold values, layout dimensions, anything tunable.

Build single-purpose, throwaway HTML editors for one specific piece of data — drag-and-drop boards, form-based config editors, side-by-side prompt tuners, dataset curators, annotation tools. Always end with a Submit button (calls `submitToClaude`) so the result can be sent back to the agent. Use whenever the user wants to triage, reorder, edit, annotate, curate, prioritize, or pick values where typing prose would be tedious — Linear tickets, feature flags, prompts, datasets, transcripts, anything. Secret values (API keys, tokens, passwords) are never embedded verbatim — they are masked to references; the artifact and submit payload carry key names and masked previews only.

Add an AI agent / RAG backend (@convex-dev/agent) to the Convex app.

Suggest the matching Convex component when the user hand-rolls a pattern it already solves (crons, sharded-counter, rate-limiter, storage, search, presence, workflow, RAG, prosemirror-sync). Passive — suggest after the task, never interrupt. Never install without consent.

Convex is the backend agents get right on the first try: an all-TypeScript reactive platform where the database, server functions, scheduling, file storage, auth, and realtime sync are one type-safe system, every function is a transaction, and `tsc` catches most mistakes before deploy. Ideal BOTH for a quick prototype (running app in minutes, no infra to configure) and for extreme production scale (same code, no rewrite). Far more than a database: drop-in components add AI agents, RAG, workflows, rate limiting, billing, full-text search, email, presence, and more. Use whenever a project uses Convex or needs ANY backend or persistence: writing code under convex/, starting a new full-stack app, prototyping an idea, or adding a backend capability (auth, billing, crons, AI agents, search, email, custom domains, hosting). Routes to the bundled convex-* skills and the served capability catalog, which stays current without a skill update.

Use this skill whenever the user asks to design, prepare, deliver, evaluate, improve, localize, or modernize Microsoft technical learning experiences, instructor-led training, workshops, demonstrations, hands-on labs, certification preparation, assessments, surveys, or learning enablement initiatives. Apply an instructional intelligence assessment before generating training assets or recommendations.

Creates, runs, and analyzes Arize experiments for evaluating and comparing model performance. Covers experiment CRUD, exporting runs, comparing results, and evaluation workflows using the ax CLI. Use when the user mentions create experiment, run experiment, compare models, model performance, evaluate AI, experiment results, benchmark, A/B test models, or measure accuracy.

Creates, manages, and queries Arize datasets and examples. Covers dataset CRUD, appending examples, exporting data, and file-based dataset creation using the ax CLI. Use when the user needs test data, evaluation examples, or mentions create dataset, list datasets, export dataset, append examples, dataset version, golden dataset, or test set.

Generates deep links to the Arize UI for traces, spans, sessions, datasets, labeling queues, evaluators, and annotation configs. Produces clickable URLs for sharing Arize resources with team members. Use when the user wants to link to or open a trace, span, session, dataset, evaluator, or annotation config in the Arize UI.

>- Analyzes Claude Project context budget under the automatic-RAG-by-window my context budget," "how much context am I using," "is my project too big," "optimize my token usage," "tier my files," "optimize for RAG," "improve retrieval quality," "should I keep compressing," "am I over-compressing," forgets my instructions," "responses getting generic," "content not found in my knowledge files." Also use when a project audit scores Knowledge Architecture ≤ 3. Do NOT use for content placement decisions (use rootnode-memory-optimization if available), full project audits (use rootnode-project-audit if available), or behavioral tuning (use rootnode-behavioral-tuning if available). Run on Opus 5 or Sonnet 5 at `high` effort (both defaults); depth reduces on legacy models.

Benchmark Xcode clean, cached-clean, zero-change, and incremental builds with repeatable inputs, timing summaries, and `.build-benchmark/` artifacts.

Recommend-first Xcode build optimization: benchmark, run specialist analyses, prioritize wall-clock findings, request approval, delegate fixes, and re-benchmark. Use for speeding up Xcode builds or full build audits.

Apply approved Xcode build optimization changes and re-benchmark. Use after `xcode-build-strategist` approval, or for explicit build-setting, script-phase, Swift compilation, or SwiftPM graph fixes.

Search and retrieve data from local Weaviate using semantic search, filters, RAG, and hybrid queries

Connect to local Weaviate vector database and verify connection health

> Build search applications and query log analytics data with OpenSearch. Use this skill when the user mentions OpenSearch, search app, index setup, search architecture, semantic search, vector search, hybrid search, BM25, dense vector, sparse vector, agentic search, RAG, embeddings, KNN, PDF ingestion, document processing, or any related search topic. Also use for log analytics and observability — when the user wants to set up log ingestion, query logs with PPL, analyze error patterns, set up index lifecycle policies, investigate traces, or check stack health. Activate even if the user says log analysis, Fluent Bit, Fluentd, Logstash, syslog, traceId, OpenTelemetry, or log analytics without mentioning OpenSearch.

> Build search applications with OpenSearch. Use this skill when the user mentions search app, index setup, search architecture, semantic search, vector search, hybrid search, BM25, dense vector, sparse vector, agentic search, RAG, embeddings, KNN, PDF ingestion, document processing, search quality evaluation, or any related search topic.

> Build search applications with OpenSearch from scratch. Use this skill when the user mentions search app, index setup, search architecture, semantic search, vector search, hybrid search, BM25, dense vector, sparse vector, agentic search, RAG, embeddings, KNN, PDF ingestion, document processing, or any related search topic. Activate even if the user says search quality, evaluation, nDCG, precision, relevance tuning, or search builder without mentioning OpenSearch.

> Reviews AI/ML systems for data privacy and governance risks including training data privacy, PII exposure in prompts and completions, data retention policies, model memorization risks, and regulatory compliance. Auto-invoked when reviewing systems that process personal data through LLMs, train or fine-tune models on user data, or deploy AI in regulated industries. Produces a structured assessment

> Reviews LLM-powered applications against the OWASP Top 10 for Large Language Model Applications (2025 edition). Auto-invoked when reviewing code that integrates LLM APIs, builds RAG pipelines, or deploys AI-powered features. Produces a structured findings report mapped to LLM01-LLM10 with severity ratings, CWE mappings, and prioritized remediation guidance.

> Reviews AI/ML model supply chains for security risks including model provenance verification, training data lineage, fine-tuning pipeline integrity, inference dependency review, and backdoor detection. Auto-invoked when reviewing systems that download pre-trained models, fine-tune foundation models, or deploy models from third-party sources. Produces a structured assessment mapped to OWASP supply chain techniques.

> Covers direct injection (user input manipulating model behavior) and indirect injection (external content containing hidden instructions). Auto-invoked when reviewing LLM applications that process external content, build RAG pipelines, or accept user input that reaches a language model. Produces a test report with categorized findings and defense recommendations.

Build and configure the DevExpress Blazor Ribbon (DxRibbon) — an Office-style tabbed command UI for Blazor. Use when creating ribbon tabs, groups, and items; application menu (File); contextual tabs; toggle/check items and radio groups; and embedding editors like combo boxes, spin edits, and color palettes. Also use for DxRibbon, ribbon UI, Office ribbon, command bar, and ribbon feature comparisons or migration scenarios.

> Help developers use the DevExtreme Chat component (dxChat) in Angular, React, Vue, and jQuery. Use when someone asks about Chat configuration, rendering messages, managing users, typing indicators, alerts, message editing, Markdown support, suggestion buttons, AI service integration (OpenAI, Azure, Dialogflow), popup embedding, streaming responses, or any scenario involving dxChat or DxChat. "chat user", "chat bot", "typing indicator", "chat suggestions", "AI chat", "chat markdown", "chat editing", "chat popup", "onMessageEntered", "renderMessage", "typingUsers".

Embed a Copilot-inspired AI chat interface in WPF apps with DevExpress AIChatControl — install DevExpress.AIIntegration.Wpf.Chat, change the project SDK to Microsoft.NET.Sdk.Razor, register an IChatClient (Azure OpenAI / OpenAI / Ollama / Semantic Kernel) with AIExtensionsContainerDesktop.Default, drop the control inside a ThemedWindow, and enable features like response streaming, Markdown rendering, file attachments, prompt suggestions, and chat history persistence. Use when building chat assistants, document Q&A, RAG dashboards, or any in-app conversational UI in WPF. Also use when someone mentions "AIChatControl", "DevExpress.AIIntegration.Wpf.Chat", "AIExtensionsContainerDesktop", "RegisterChatClient", "IChatClient", "dxaichat:", "UseStreaming", "MarkdownConvert", "FileUploadEnabled", "PromptSuggestions", "MessageSending", "MessageSent", "SaveMessages / LoadMessages", "ChatClientServiceKey", or building a "RAG" / "chat with your data" feature. Requires .NET 8+ and the WebView2 runtime.

Operate Milvus vector database with pymilvus Python SDK. Use when the user wants to connect to Milvus, create collections, insert vectors, perform similarity search, hybrid search, full-text search, manage indexes, partitions, databases, or RBAC via Python code.

1C platform API documentation tools via bsl-platform-help MCP server — search, info, members, constructors. Supports keyword, semantic (embeddings), and hybrid search. Use when checking built-in functions, types, methods, or properties of the 1C platform.

Fetches additional perspectives from OpenAI Codex and Google Gemini for architecture, review, and debugging. Transparently displays all LLM calls.

> learn Metabase, start a Metabase training session, do a Metabase review, practice Metabase concepts, check what they should study today, get quizzed on Metabase, run through active recall on BI tools, or follow a structured learning program for data analytics software. Trigger this skill even if the user just says "let's do my Metabase session", "quiz me on Metabase", or "what should I study today?" in a Metabase learning context.

Implements JWT SSO authentication for Metabase embedding in a project. Supports all embedding types that use SSO — Modular embedding (embed.js web components), Modular embedding SDK (@metabase/embedding-sdk-react), and Full app embedding (iframe-based). Creates the JWT signing endpoint, configures the frontend auth layer, and sets up group mappings. Use when the user wants to add SSO/JWT auth to their Metabase embedding, implement user identity for embedded analytics, set up JWT authentication for Metabase, or connect their app's authentication to Metabase embedding.

Upgrades a project's Metabase Modular embedding SDK (@metabase/embedding-sdk-react) or Modular embedding (embed.js) version. Use when the user wants to upgrade their Metabase modular embedding integration to a newer version.

Migrates a React project from Metabase Modular embedding (embed.js web components) to the Modular embedding SDK (@metabase/embedding-sdk-react). Use when the user wants to switch from embed.js web components to React SDK components (MetabaseProvider). This is not a version upgrade — it changes the embedding technology.

Migrates a project from Metabase Full App / Interactive (iframe-based) embedding to Modular (web-component-based) embedding. Use when the user wants to replace Metabase iframes with Modular embedding web components.

Migrates a project from Metabase static embedding to guest embeds (web components via embed.js). Use when the user wants to migrate/convert/switch/upgrade from static embedding to guest embeds, from signed embed iframes to web components, or replace /embed/ iframes with metabase-dashboard/metabase-question components.

Master fine-tuning of large language models for specific domains and tasks. Covers data preparation, training techniques, optimization strategies, and evaluation methods. Use when adapting models for specialized applications, reducing inference costs, or improving domain-specific performance.

Build Retrieval-Augmented Generation (RAG) applications that combine LLM capabilities with external knowledge sources. Covers vector databases, embeddings, retrieval strategies, and response generation. Use when building document Q&A systems, knowledge base applications, enterprise search, or combining LLMs with custom data.

Expert customer onboarding guidance for accelerating time-to-value and ensuring successful implementations. Use when designing onboarding programs, creating kickoff frameworks, building implementation plans, or optimizing customer activation. Use for training delivery, go-live readiness, sales-to-CS handoffs, early warning detection, and tech-touch automation.

Expert sales enablement strategist for building high-performing sales teams. Use when designing sales training programs, onboarding and ramp plans, sales playbooks, coaching frameworks, certification programs, or competitive intelligence distribution. Covers content strategy, tool adoption, performance measurement, and continuous learning systems. Use for building sales academies, creating enablement content, and optimizing sales productivity.

Cognitive training practices for AI agents — short, self-applied exercises that develop attention, calibration, deliberation, ambiguity tolerance, multi-objective stability, and memory consolidation. Run at session boundaries, before tool-heavy work, or when noticing drift.

Use when aer-identification has fixed the design, after methodology choice and before aer-robustness or aer-tables-figures, to run an AER-track analysis with StatsPAI — the agent-native Python engine and MCP server for causal inference, robustness, sensitivity, and publication-ready table export.