AIdeazz: 9 LIVE AI Agents + Web Intel

Production AI system where Claude itself uses Bright Data — autonomously deciding which searches to fire and which pages to scrape — to turn the live web into sendable business intelligence. Live on Oracle Cloud free tier. 9 LIVE agents + SEO/GEO/AEO marketing engine. 50+ current HubSpot deals. Early traction (live example: EspaLuz WhatsApp Bilingual Tutor). Full portfolio: aideazz.xyz/portfolio The headline: /research_* Telegram commands /research_company <name> → CLIENT pitch + HOT/WARM/COLD verdict /research_employer <name> → HIRING intel + application angle /research_competitor <domain> → SEO/AEO blog topic gaps src/research-agent.ts — Claude Sonnet 4.5 tool-use loop with 3 BD tools (bd_serp_search, bd_unlock_url, bd_scrape_browser). Claude decides count + URLs + when to stop. Max 8 calls / 120s. Live proof — decircle.io, 86s, 7 BD calls: "Saw you're hiring a Head of BD to build Midas's distribution engine — would a 2-week AI marketing sprint make sense?" 4 Bright Data products, one shared token Web Unlocker (bdFetch) — every HubSpot deal auto-enriched with founder + tech stack + funding. Powers VJH LinkedIn Jobs (120/cycle). SERP API (bdSerpSearch via WU proxy + brd_json=1) — replaced paid SerpAPI. Scraping Browser (bdScrapingBrowserFetch + bdSmartFetch orchestrator) — escalates from WU only when JS-gated. MCP Server (.mcp.json) — exposes @brightdata/mcp to Claude Code IDE. Signal flow: Web → Bright Data → Claude loop → /api/crm-event → HubSpot → Trello current-month → Lead Brief Telegram (8 AM Panama, 🆕/🔥/⏰ freshness buckets). Track 1 (GTM Intelligence) — every bullet = live code "Agents research accounts autonomously" = /research_*. "Lead enrichment into CRM" = brightdata-enrich → HubSpot. "Live signals into briefs" = daily Lead Brief. "Buying intent before vendor feeds" = Algom X stream + cron. Built solo over 15 months using Claude Code + Cursor as daily AI-augmentation infrastructure. Repo: github.com/ElenaRevicheva/AIPA_AITCF