9 autonomous agents in production + AILA blueprint. One developer since Jan 2025. Under $15,000. The stack is still shipping — gates, wiki, outreach, concierge — on the same shared infrastructure.
Built to prove I can ship. Primary goal: join the right team as a founding AI / applied-AI engineer. Investment is welcome — stable income is the mission.
Especially when you are adapting to a new country, a new language, or a new career — without a team, a budget, or a safety net.
280M+ people navigating daily life in a second language with no affordable, always-on support.
Job hunting rewards templates, not signal. ATS forms, ghosting, zero feedback loops — especially for non-traditional profiles.
Founders and small teams can't afford the engineering bandwidth that their ideas actually require.
Not a platform. Not a wrapper. Each agent is a production system: scraping, scoring, deciding, acting, notifying — 24/7. WhatsApp send is still a human click. Blog, wiki, email, and research loops are gated and logged.
All agents share one Oracle Cloud VM and one Oracle Autonomous Database. A watchdog polls every product's health check and restarts what dies — failures surface in Telegram, never silently. ~$0/month on Oracle free credit grant.
Five-provider fallback chain — Claude, OpenAI, Gemini, Grok, Groq — ordered per task: quality paths lead with the strongest model, classification with the cheapest that clears the bar. Provider retirements are a one-line config change.
Groq Whisper transcription. Accessible via WhatsApp and Telegram — no app install required. Sub-2s latency on EspaLuz.
550-test, 4-layer VJH harness ($0.03/run). Daily blog grounding gate: fail-closed on unsourced numbers, fail-open on cadence. Named failure modes live in the AI Ops Wiki (21 incidents, 18 failure modes).
Every agent below runs 24/7 on Oracle Cloud, coordinating through shared database tables. Verifiable via pm2 list + systemctl is-active. Real users. Real deploys. AILA is the next-layer architecture blueprint, not yet deployed.
| Agent | Function | Platform | Status |
|---|---|---|---|
| EspaLuz WhatsApp | Bilingual tutor, voice-first, 50+ emotion detection | LIVE <20 paying | |
| EspaLuz Telegram | Bilingual tutor, structured lessons | Telegram | LIVE |
| EspaLuz Influencer | Automated social content EN/ES | Telegram | EARLY |
| CTO AIPA | Ops brain — GitHub webhook → diff → LLM review → comment; outreach, concierge, briefings | Telegram + API | LIVE 80+ commands |
| CMO AIPA | GEO+SEO, grounded daily blog, wiki, inbound concierge, outreach | Oracle + HubSpot + Telegram | LIVE gated |
| Atlas | Ad-market radar + campaigns — detect → sell → measure, weekly, 7 verticals, GA4 outcomes fed back | Web + Telegram + HubSpot | LIVE weekly |
| VibeJob Hunter | Autonomous job search, scoring, ATS automation | Telegram | VERIFYING |
| OpenClaw Shortlist | YC AI company shortlist, voice interface | Telegram | LIVE |
| Algom Alpha / DragonTrade | Crypto research + auto-posting | X (Twitter) | FRAGILE |
| Atuona Creative AI | NFT gallery + automated AI film studio — clips in, finished film out | Polygon + Telegram + Web | LIVE 6 films |
| AILA | Adaptive Intelligent Life Assistant — orchestrates all 9 agents, longitudinal personal memory | Hive / Oracle | BLUEPRINT |
Fewer than 20 paying subscribers — but they have been retained since 2025 with zero marketing spend. People paid, stayed, kept using it. That is the retention signal worth scaling.
The technical co-founder as software: /code and /fix open real pull requests from a phone, a human keeps the last click, and a synthetic lead runs the whole pipeline daily at 07:45 — silent on pass, alerting on failure. Multi-model routing with explicit cost reasoning.
7-node StateGraph: gate → score → route → submit/outreach/discard → notify. SQLite checkpointer per job_id eliminates duplicate processing. Human-approval interrupt for score 60–69. Real HTTP status + confirmation_id capture on every ATS submission. First Oracle cycle: 8 jobs, all correctly discarded, zero errors.
Sep 2026: the target-role list lives in one shared registry rendered into every prompt and gate, guarded by 413 contract tests; a dry-run replay of 51 rejected jobs proved 20 were wrong before deploy.
JSON-LD, OpenGraph, hreflang EN/ES, build-time sitemap. Money page is /portfolio — every surface links back. Since Sep 2026, llms.txt and a dated geo-manifest also point answer engines at the Visibility API and the podcast.
Own blog at aideazz.xyz/blog + Dev.to canonical. Unsourced numbers skip. Cadence salvages from measured evidence so the day still ships.
Stops near-copies, not just identical slugs. Existing duplicates get a canonical — never deleted. Never canonicalise onto a fabricated-stack post.
HubSpot chat + portfolio form → Telegram → Claude Fable 5 drafts the reply → Resend. A dedicated watchdog re-drafts through the 5-provider chain if the primary dies, so one vendor outage does not kill the inbox.
Prospects discovered automatically (SERP + hiring boards + product launches, Bright Data enrichment) land in HubSpot with a tested WhatsApp link and a one-click email. A send moves the deal to Sent and creates a follow-up task; Resend webhooks mark delivered / opened on the deal — true delivery data, not assumptions.
Named failure modes from production — including the prompt is a source. Live at /ai-ops-wiki.html.
Public REST API. 34 automated checks in four weighted categories — AI crawler access 25%, structured data (GEO) 25%, answer-readiness (AEO) 30%, technical foundation 20% — returning a 0–100 score, a grade and a prioritised fix list.
Reports separately whether GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended and CCBot may read the site — the blocker most owners never see.
Instant check at aideazz.xyz/api, or call it with the public demo key. The API is also the opening offer for AEO/GEO consulting work.
AI podcast by Elena Revicheva — 11 episodes live, with an RSS feed. Listen at podcast.aideazz.xyz.
English and Spanish at aideazz.xyz/blog, cross-posted to Dev.to. A post that cites a number it cannot source does not publish.
Real production incidents, each with its named failure pattern, written from verified records — never re-told by a model. /ai-ops-wiki.html
HOW THEY CONNECT — build sessions become wiki incidents → the wiki publishes to the blog and Dev.to → all 11 podcast episodes link back to their blog posts → blog and wiki are listed for AI crawlers in llms.txt, the sitemap and a dated GEO manifest → a weekly probe checks whether AI answer engines actually cite them.
JSON-LD for the organisation, the person, the site, FAQs and lists; llms.txt plus its .well-known mirror; a dated geo-manifest naming the API, blog, wiki and podcast.
150+ grounded blog posts (EN/ES), 21 wiki incidents with named failure modes, 11 podcast episodes linked to their posts, FAQ markup on the ops runbook — every figure traceable to a source.
robots.txt explicitly allows AI crawlers (GPTBot, ClaudeBot, PerplexityBot and more); a 173-URL sitemap plus the podcast sitemap; static pages served from an IPFS edge.
aideazz.xyz 100 A+ · podcast 100 A+ · ops runbook 100 A+ · /api 98 A+ · wiki 93 A+ · atuona.xyz 69 C — 0 of 6 AI crawlers allowed: the audit's own finding, fix queued.
LinkedIn, Instagram and YouTube through Buffer and Make.com — including a daily rotation of the API's own product film — and X through Algom Alpha on the X API.
A weekly probe asks Google AI Overviews, Gemini, OpenAI search and Perplexity real buying questions and records whether AIdeazz is cited, named without a link, or missed.
May was the Bright Data loop. June–August was gates, a wiki, and making sure a publish never claims a resume it did not attach. Same VM. Same 9 agents. Better engineering.
Daily publisher refuses unsourced numbers. A leftover figure in the prompt is treated as a source — that is a named failure mode. Salvage rewrites against measured evidence; if that still fails, the day ships an evidence-only post. Silence is correct only when there is nothing to measure.
Production incidents, one concept each — including the prompt is a source and git is not the origin (the live IPFS pin, not the git SHA, is what the public site is). No customer data. Verified from logs.
Resend MIME attach. PDFs only, from one folder, ≤5 MB, %PDF- magic-byte checked. If a listed file fails to load, the send is refused — the email never claims a resume it did not attach.
Portfolio form and HubSpot chat draft via Claude Fable 5, with a watchdog that re-drafts through a 5-provider chain if the primary dies. Near-duplicate blog posts are blocked by token overlap, not slug equality. Old copies get a canonical tag — never a delete.
Claude Code and Cursor are the actual development environment — not a shortcut, not training wheels. The permanent workflow.
Every agent was scoped, built, and shipped to production by one person. 9 systems + AILA blueprint since Jan 2025 is the proof.
No black boxes. No delegated understanding. Every line is reviewed, every trade-off is deliberate.
Not in staging. Not in demos. Real WhatsApp subscribers, real GitHub webhooks, real ATS submissions.
Multi-model routing. Free-tier APIs where quality allows. ~$0/month operating cost during credit grant.
Deployed on Polygon during the 2025 build cycle. All contracts verifiable on-chain. Included for completeness — not as the value proposition.
48+ NFTs minted via Atuona Creative AI — automated image + video generation pipeline.
Local credibility. Access to Panama government and business leaders. Regional ecosystem support.
Individual member. Global AI agent community. Industry recognition for independent builders.
The work is real. The systems are live. The honest goal is stable income — joining a worthy team first, investment second.
Full-time remote, founding / senior AI or applied-AI engineer — or AI Product Manager, AI Solutions Architect, AI Systems Consultant, fractional Chief AI Officer. $3.5K–$5K+/mo USD. LATAM-friendly hours. Someone who can design, build, and ship AI systems end-to-end — board-level operator + production builder. Rare combination, fully verifiable above.
Fix EspaLuz payment tracking. Verify VibeJob Hunter delivery. Hire one engineer. Extend infra runway before Oracle credit grant expires. Small marketing budget to test retention at scale.