<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>ML Engineering Articles: Search, RecSys, LLM Agents, MLOps</title><description>Production ML engineering articles on Search, RecSys, LLM agents, and MLOps: architecture decisions, release patterns, evaluation methods, and reliability practices.</description><link>https://igor-ya.com/</link><language>en-US</language><item><title>Evals for LLM Agents: The Minimal Production Set</title><link>https://igor-ya.com/posts/llm-agent-evals-production-framework/</link><guid isPermaLink="true">https://igor-ya.com/posts/llm-agent-evals-production-framework/</guid><description>The minimal production eval set for LLM agents: outcome and trajectory evaluation, code and model graders, pass^k reliability, judge calibration, release gates.</description><pubDate>Tue, 09 Jun 2026 00:00:00 GMT</pubDate><category>LLM</category><category>Agents</category><category>Evals</category><category>LLM-as-judge</category><category>AgentOps</category><category>Observability</category><category>CI/CD</category><category>Release Gates</category></item><item><title>Search and Recommendation Logs as Data for LLM Post-Training</title><link>https://igor-ya.com/posts/product-logs-post-training-llm-search-recsys/</link><guid isPermaLink="true">https://igor-ya.com/posts/product-logs-post-training-llm-search-recsys/</guid><description>How to turn search and recommendation logs into datasets for LLM post-training: SFT, DPO, GRPO, hard negatives, LLM-as-judge, and release gates.</description><pubDate>Thu, 28 May 2026 00:00:00 GMT</pubDate><category>LLM</category><category>Post-training</category><category>SFT</category><category>DPO</category><category>GRPO</category><category>LLM-as-judge</category><category>Search</category><category>RecSys</category><category>MLOps</category><category>Data Infrastructure</category></item><item><title>Multimodal Retrieval for LLMs</title><link>https://igor-ya.com/posts/multimodal-retrieval-llm-context-selection/</link><guid isPermaLink="true">https://igor-ya.com/posts/multimodal-retrieval-llm-context-selection/</guid><description>How multimodal retrieval is used around LLMs: hybrid search, visual document retrieval, reranking, context packing, citations, long context, agentic search, and eval.</description><pubDate>Sat, 25 Apr 2026 00:00:00 GMT</pubDate><category>Multimodal Retrieval</category><category>LLM</category><category>Search</category><category>RAG</category><category>Visual Document Retrieval</category><category>Reranking</category><category>Context Engineering</category><category>GPT</category><category>Claude</category><category>Gemini</category></item><item><title>Adding Tool Calling to Search Systems Without Breaking Retrieval, Reranking, or Control</title><link>https://igor-ya.com/posts/agentic-search-production-tool-calling-retrieval-reranking-control/</link><guid isPermaLink="true">https://igor-ya.com/posts/agentic-search-production-tool-calling-retrieval-reranking-control/</guid><description>A production guide to placing tool calls before retrieval, after reranking, or after answer selection without losing relevance, latency, safety, or rollback control.</description><pubDate>Sat, 21 Mar 2026 00:00:00 GMT</pubDate><category>Search</category><category>Retrieval</category><category>Reranking</category><category>Tool Calling</category><category>Agents</category><category>MCP</category><category>Observability</category><category>AI Security</category></item><item><title>Assistants API to Responses API Migration: Production Playbook Before August 26, 2026</title><link>https://igor-ya.com/posts/assistants-api-to-responses-api-migration-playbook-2026/</link><guid isPermaLink="true">https://igor-ya.com/posts/assistants-api-to-responses-api-migration-playbook-2026/</guid><description>A production migration guide from Assistants API to Responses and Conversations covering timeline, entity mapping, breaking changes, rollout strategy, and parity testing.</description><pubDate>Tue, 03 Mar 2026 00:00:00 GMT</pubDate><category>Assistants API</category><category>Responses API</category><category>Conversations API</category><category>OpenAI</category><category>Migration</category><category>AgentOps</category><category>MLOps</category><category>LLMOps</category><category>AI Engineering</category></item><item><title>Offline-Online Gap in RecSys: 11 Release Gates and Incident Playbook</title><link>https://igor-ya.com/posts/deep-learning-recsys-offline-online-gap-production/</link><guid isPermaLink="true">https://igor-ya.com/posts/deep-learning-recsys-offline-online-gap-production/</guid><description>A production guide to RecSys offline-online failures: feedback loops, delayed labels, train-serve skew, OPE limits, release gates, and incident response playbooks.</description><pubDate>Thu, 26 Feb 2026 00:00:00 GMT</pubDate><category>Deep Learning</category><category>RecSys</category><category>MLOps</category><category>Feedback Loops</category><category>Delayed Feedback</category><category>Feature Skew</category><category>Counterfactual Evaluation</category><category>Observability</category></item><item><title>Agent or Workflow: How to Choose Architecture Without Hype</title><link>https://igor-ya.com/posts/agent-vs-workflow-architecture-framework/</link><guid isPermaLink="true">https://igor-ya.com/posts/agent-vs-workflow-architecture-framework/</guid><description>A practical framework for deciding between workflow automation and agent architecture, including safety boundaries, eval design, cost trade-offs, and rollout guidance.</description><pubDate>Wed, 18 Feb 2026 00:00:00 GMT</pubDate><category>LLM</category><category>Agents</category><category>Workflow</category><category>System Design</category><category>AgentOps</category><category>Evals</category><category>AI Security</category><category>FinOps</category></item><item><title>MLOps for a Support RAG Agent in 2026: Releases, Security, and Cost</title><link>https://igor-ya.com/posts/mlops-rag-agent-support-release-gates-security-cost-2026/</link><guid isPermaLink="true">https://igor-ya.com/posts/mlops-rag-agent-support-release-gates-security-cost-2026/</guid><description>A production guide to shipping a support RAG agent with release gates, policy boundaries, tracing, evaluation loops, and cost control.</description><pubDate>Tue, 10 Feb 2026 00:00:00 GMT</pubDate><category>MLOps</category><category>RAG</category><category>AgentOps</category><category>LLMOps</category><category>AI Security</category><category>Observability</category><category>FinOps</category></item><item><title>MLOps for Production ML: 7 Release Gates for Controlled Rollouts</title><link>https://igor-ya.com/posts/mlops-release-gates-production-ml/</link><guid isPermaLink="true">https://igor-ya.com/posts/mlops-release-gates-production-ml/</guid><description>A production MLOps guide to the seven release gates that keep model rollouts inside quality, latency, reliability, and cost limits.</description><pubDate>Fri, 26 Dec 2025 00:00:00 GMT</pubDate><category>MLOps</category><category>Model Registry</category><category>CI/CD</category><category>Observability</category><category>Drift Detection</category><category>FinOps</category><category>SRE</category><category>AI Security</category></item><item><title>igorOS: A Browser-Based Agent Interface You Can Actually Use</title><link>https://igor-ya.com/posts/igoros-alternative-site/</link><guid isPermaLink="true">https://igor-ya.com/posts/igoros-alternative-site/</guid><description>A browser-based agent interface demo that shows how tool calling, app state, and visible execution loops work inside a desktop-style environment.</description><pubDate>Mon, 15 Dec 2025 00:00:00 GMT</pubDate><category>Web OS</category><category>Tool Calling</category><category>Agent UX</category><category>React</category><category>TypeScript</category><category>UI</category></item><item><title>Training a Hybrid LLM and Recommender System with Semantic IDs</title><link>https://igor-ya.com/posts/semantic-ids-llm-recsys/</link><guid isPermaLink="true">https://igor-ya.com/posts/semantic-ids-llm-recsys/</guid><description>How to train a hybrid LLM and recommender system with semantic IDs, retrieval-aware objectives, and controllable recommendation outputs.</description><pubDate>Mon, 20 Jan 2025 00:00:00 GMT</pubDate><category>LLM</category><category>Recommendations</category><category>Semantic IDs</category><category>RQ-VAE</category><category>Qwen3</category><category>Retrieval</category><category>Ranking</category><category>SASRec</category></item></channel></rss>