Public work, not slide decks.

A 12-agent, 7-wave research pipeline producing a typed, source-cited knowledge graph of 100 deployable AI/digital-government initiatives across Uzbekistan and Kyrgyzstan, plus 200 solopreneur-MVP ideas grounded in the same graph. The graph maps every initiative to the specific decree, institution, decision-maker, donor programme, and global precedent it depends on. 882 records, 100 decrees, 105 institutions, 117 named decision-makers, 49 donor programmes, 50 live tenders — every numeric claim is source-cited or marked not_found. A separate 16-specialist AI Audit Team (information architect, mobile-first QA, AI-search optimiser, accessibility, link-verifier, etc.) re-audits every page on every build.

Why this matters. It is the proof point for "ship investment-grade market research from a single prompt." The repo is the working demonstration of multi-wave agent orchestration, cross-model verification on a strict $20 OpenRouter budget, and a 14-check content + SEO quality gate. Live country reports for Uzbekistan and Kyrgyzstan are now public.

An eight-page strategic brief on padel coaching technology — competitor landscape, subscription economics, MVP design, and a 90-day operating plan. Every numeric claim has a verifiable source URL. The pipeline that built it runs under Claude Code and OpenRouter, fans out across multiple frontier models, and gates output through a 14-check unified content + SEO quality script before merge.

Why this matters. The earlier reference implementation for everything else here. It demonstrates a working multi-agent orchestration pattern, an evidence-traceability discipline, and a mobile-first, AI-search-optimised publishing standard — all three of which carry directly into product work.