Alex Maldonado · Los Angeles · AI portfolio, October 2026
Marketing systems that run while I sleep.
I build AI pipelines that research, write, design, publish, follow up and report on a schedule, then I measure whether they worked. This page is the record: what each system does, how it is wired, and the numbers it produced.
Autonomous content pipelines
Each of these runs without me touching it. Most are scheduled agents or GitHub Actions jobs that pull a source, write, render, publish through Metricool or WordPress, and log what they did.
AI Twin daily video
A private repo renders one short video of my own AI avatar every day and posts it to Instagram Reels, TikTok and Threads for my @the.alexmaldonado personal brand. Scripts are generated from a trend feed (Google News, a YouTube channel, an RSS feed), voiced in my cloned voice, and lip-synced onto one of 15 looks rotated in order.
- Rebuilt on the supported API after the old Captions endpoints died; audio goes in, video comes out, so voice synthesis moved to ElevenLabs.
- Cost guard: the run refuses to render twice in one day.
- Slot collision diagnosed when TikTok spam-flagged a post: two jobs shared 18:00, so the video got its own hour. AI-content flags are set on Instagram and TikTok.
Threads to TikTok carousel
Recycles my best-performing Threads posts of the fortnight into a branded TikTok and Instagram photo carousel, for both the real estate and personal accounts. Posts beating the 75th percentile are selected, capped at 12 slides; a ledger on the repo’s media branch guarantees no carousel repeats.
- Filters out templated autolist filler so recycled content is the real stuff, and dropped a third-party image host for the repo’s own media branch when it started failing.
Daily blog publishing, two sites
Keyword targets come from SE Ranking, drafts are written in the blog’s voice, a Pexels banner is uploaded with alt text, Yoast focus keyphrase, title and meta description are set through the REST meta API, internal links are added, and the post publishes live. A trend-driven post on AI and marketing news runs alongside the evergreen SEO posts.
- Retro SEO pass over 48 existing posts in one session: new banners, rewritten Yoast fields, deduped headings, recategorized posts, “Related reading” blocks, and contextual in-body links averaging 7 unique destinations per post.
- Five thin posts rewritten from 356 to 910 words with zero H2s into 857 to 930 words with 6 to 8 H2s, FAQ blocks and snippet paragraphs.
- Service pages built through the API by writing Elementor data directly, including three vertical pages live the same day.
Daily LinkedIn, three surfaces
Three posts a day: my real estate profile, my marketing profile and the Maldonado Marketing company page. Each has a weekday rotation of content pillars and CTAs (DM, comment a keyword, link, soft question). Posts queue a day ahead so I have an overnight window to edit or kill one.
- Guardrail: never invent a statistic. Every figure is checked against Freddie Mac, Zillow or C.A.R. that run and cited with a date, or the post ships without numbers.
Branded daily news graphics
Two scheduled posts, one for the agency on trending marketing and AI news for law firms and small businesses, one for the real estate brand on a trending housing topic. Each researches social and news, composes a branded graphic with a Pexels photo and the key text inside it, and posts across every connected account.
Engagement research lists
Each morning an agent finds the posts worth replying to in my niches and drafts a comment for each using the NEPQ question-led sales framework. The list goes to a VA to post, so the account stays human-operated while the research and copy are automated.
Lead generation and follow-up agents
Outbound and nurture that would normally take a sales team. Approval gates sit where a mistake is expensive; everything else is automatic.
Always-on cloud agent
A persistent agent I talk to through a Telegram bot. Its standing jobs: find LA and Orange County businesses whose Google listing has no website, build a free mockup, and text the offer; apply to remote marketing roles that fit my criteria; and take over any scheduled task that fails to run on its own.
- Pipeline behind it: 159 no-website businesses found, deduplicated against 903 existing CRM records with zero collisions; yield measured by vertical (tailors 75 to 85 percent with no site, upholstery 65, auto detailing 55, landscaping 5).
- Delivery fault found and fixed: 79 percent of texts were arriving as a carrier fallback notice because anything over about 305 characters was silently converted to MMS. Delivery went from 21 to 73 percent on the same audience.
- Refused the cold-text version as first requested on TCPA and carrier-filtering grounds, and built the call-first version instead.
Ulinc re-engagement and NEPQ replies
Every Monday, real estate leads with six or more months of silence are enrolled in a four-step LinkedIn nurture. Every morning, replies across both Ulinc accounts are drafted using NEPQ and emailed to me for approval, with the goal of a booked call. People who showed interest and went quiet get weekly follow-ups, capped at four, with a check that nothing was sent recently.
RealScout listing texts
Reads each client’s RealScout alert emails and saved-search criteria, picks their two or three best-fit listings by budget and needs, and texts them through Follow Up Boss. Rentals included. A weekly “This Week Around L.A.” client email is staged as a draft in the same CRM so I review and press send.
auditvid
A command-line tool that researches a B2B prospect, writes a personalised audit-video script, renders it with my avatar, and outputs a Ulinc import CSV. Priorities in order: audit specificity, cost control, throughput. Projected cost is shown before any batch runs.
Apollo outbound rebuild
Rebuilt the agency’s cold-email program on a separate sending domain so the primary domain never touches outreach: dedicated Workspace, SPF and MX live, a three-step sequence for 1 to 50 employee injury firms, and a verified list of 60 owner and managing-partner contacts from a pool of 5,798, with no role-based or catch-all addresses.
Chief-of-staff morning brief
One email every weekday that reads across six systems and tells me what needs a decision today. It is the layer that lets the other pipelines run unattended.
B&D Injury Law Group, Marketing Director
In-house since July 2026 for a five-office personal injury firm with a bilingual site and a roughly $1M ad budget. I lead a team of six and the Black Diamond Law rebrand. The first job was not growth: it was making the site technically sound and the lead data true. Everything below is logged, dated, on a monday.com work record built as proof of work.
Lead attribution system
Website form leads had never been measured. I built first-touch capture (a 90-day cookie plus per-visit page trail), hidden fields on all four site forms, matching GA4 custom dimensions, a server-side lead-source table with a ranked “Lead Sources” screen inside wp-admin, and tiered attribution (content, channel, baseline). The firm’s lead names and case details are deliberately never stored.
- Found a sitewide fault: the footer forms were sending frozen source data, so every footer lead reported the same blog post. Fixed client side and again server side so it cannot silently regress.
- Weekly attribution record republishes itself from a key-gated REST route that returns aggregate counts only, with a valuation calculator for the year-end case.
- wp-admin Command Center: replaced thirteen purposeless dashboard widgets with traffic, blog pipeline, leads-by-page and content-health panels.
Technical SEO remediation
Full site cleanup after a redesign, with a weekly SEO data board as the single source of truth so reports read from stored data instead of re-pulling the API. Keyword tracking rebuilt around the firm’s five markets (LA, Las Vegas, Bakersfield, San Francisco, Orange County), Spanish terms and high-value case types.
Content engine and Spanish mirror
Every post published goes onto a partner approval board with a live link and an Approved / Needs Changes status. A blog engagement layer (quick-answer boxes, in-article navigation, related posts) was written to 29 posts and render-verified across all 117 in the sitemap. The whole English site, blog excluded, now has a Spanish counterpart with Spanish-to-Spanish internal links.
- Advertising compliance standard written from primary sources after SB 37 amended the Business and Professions Code, and wired into the content pipeline.
Local authority and PR program
Link building through usefulness. A 422-organization community board (humane societies, CPR trainers, neighborhood councils, Rotary clubs, chambers, nonprofits across all five markets) and a 161-item legal directory board, every name and URL verified. Each pitch offers a specific resource for a specific page the recipient already runs: a Good Samaritan explainer for an AED program, a pedestrian guide for a neighborhood council’s links page, a free know-your-rights talk for a day labor program.
- Media: weekly podcast and press pitches for the firm’s intake director (Law.com, SiriusXM, AARP, HerMoney, Law&Crime), each sent as a Gmail draft for approval before anything goes out.
- Reporting: daily status to partners, weekly status with the numbers, weekly intake reports (share of calls answered live), SE Ranking credit burn alerts, and a fellowship outreach board for law schools and bar groups.
- Reputation: cleared a backlog of 15 unanswered four and five star reviews in one session, four in Spanish, with compliant reply libraries for every location.
Black Diamond Law rebrand
Logo system (black diamond, shield, scales, gold on black), a résumé-style credibility site with case results and a Spanish switch, and a free case review funnel that scores an inquiry against the firm’s own criteria, routes it to qualified, manual review or declined, and hands staff a dashboard with CRM sync status. AI-generated 15-second TV spots and a 57-image social batch produced in-house.
Client builds and in-house products
A selection from the 27 live builds on maldonadomarketing.co/portfolio. The ones below were built on an AI-assisted app stack with real databases, dashboards and tracking behind them.
Mount Olive Family Services
Full stack in Spanish for a licensed insurance agent: lead site with an insert-only database, passcode-gated mobile leads dashboard, GoHighLevel CRM built through the API (10 custom fields, 14 tags, three pipelines, Spanish calendar), and ten Spanish nurture emails. Caught a silent API fault that returned success while dropping the consent timestamp. Three requested tactics declined on insurance advertising grounds and rebuilt.
Visit the siteBlack Diamond Free Case Review
Multi-step qualifier that scores inquiries, routes them, and gives staff verification records and exports.
Open the funnelGeoola & Associates
California and Nevada trial firm site with GA4 wired into an internal dashboard showing which parts of each page people touch.
Visit the siteInnerlings
Gamified habit tracker where a companion grows as habits stick. Accounts, streaks and progress persist in a database; installable as an app.
Open the appG.O.D. Glory Over Darkness
Print-on-demand streetwear with card checkout and an admin dashboard for orders, variants and shipping rules.
Visit the storeAlex Maldonado Real Estate
Buyer and seller guides, a Spanish section with its own hreflang, gated PDF guides, and a 404 handler redirecting 5,123 retired listing URLs. Impressions went 63 to 231 the week seven neighborhood guides went live; two authority links earned at no cost.
Visit the siteOriginal research
Findings I published rather than claims I made.
Stack
Agents and models
- Claude (scheduled tasks, Chrome, API)
- OpenClaw on Ampere.sh
- GitHub Actions
- ElevenLabs voice clone
- Mirage / Captions avatars
- invideo, Seedance
Publishing
- WordPress REST, Elementor, Yoast
- Metricool API
- Lovable (React, Tailwind)
- Gamma, Canva
- Pexels
Data and tracking
- GA4, Search Console, Looker Studio
- SE Ranking API
- monday.com (11 boards, guest seats)
- WPCode server-side logging
- IndexNow, Redirection
CRM and outreach
- GoHighLevel API
- HubSpot, Follow Up Boss
- Apollo.io, Ulinc, Sales Navigator
- SimpleTexting, Telegram
- Qwoted
How I work with AI
Automate the loop, gate the risk
Content, research and reporting run fully unattended. Anything that sends a message to a real person, spends money, or touches a licensed professional’s compliance waits for a yes.
Find silent failures by measuring output
An API returning success while dropping a legal record. Texts replaced by a carrier notice. Form data frozen sitewide. None of these throw an error. They show up only if you check the result.
Verify at source, report what is true
Eight hundred link opportunities reported as eighteen. A county eligibility claim checked nine times before it went into an article. Smaller and correct beats larger and wrong.
Compliance is part of the deliverable
California attorney advertising rules, Insurance Code 787, TCPA and carrier policy, real estate advertising rules. I will say no, then build the version that works.
Log everything, including mistakes
Every task, email and document is dated on a work record. Wrong theories are recorded as disproved rather than quietly replaced.
Bilingual by default
Spanish is written, not translated, with real hreflang, Spanish keyword research and Spanish intake forms.
Want a system like this running in your business?
Start with a free audit. I look at your site, ads and follow-up before we talk, and you leave the call with a plan for what to fix first.
