An engineering guide to turning product logs from search, recommendations, and multimodal retrieval into data for SFT, DPO, GRPO, reward models, and LLM-as-judge pipelines.
Articles
Published 10 articles on Search, RecSys, LLM agents, and MLOps: architecture, quality, reliability, and cost in production.
An engineering article on the context selection layer around GPT, Claude, and Gemini: how to search, rank, and package PDFs, tables, screenshots, and visual evidence for grounded LLM answers.
A field guide for search teams adding tool calling as a bounded capability layer without losing relevance, latency discipline, safety, or rollback control.
Official timeline, architecture mapping, breaking changes, and a practical runbook to migrate from Assistants API to Responses plus Conversations without service regressions.
A practical guide to offline-online regressions in RecSys: feedback loops, delayed labels, train/serve skew, OPE limits, 11 release gates, and an incident playbook.
A practical engineering framework for choosing between workflow and agent: criteria, architecture patterns, evals, security, cost, and rollout plan.
A practical guide to shipping a support RAG agent with tool-calls: architecture contract, release gates, policy enforcement, observability, and FinOps.
A practical MLOps framework for model releases: which gates are mandatory before rollout, and how to keep quality, SLO, and cost under control.
igorOS is a browser-based desktop demo that shows what agent UX looks like when a model can call tools, manipulate state, and act inside a visible environment.
How to teach a language model to understand a catalog through semantic IDs and produce controllable recommendations with explanations