STATUS: ONLINE ▮ // OPERATOR: O.BENAVIDES // BASE: AUSTIN, TX // CLASS: BUSINESS OPERATOR × SYSTEMS BUILDER
A business person who ships.
Octavio Benavides
I find real problems inside my own businesses and build the systems that solve them, then I keep them running.
I'm a finance senior at UT Austin's McCombs School of Business. This summer I underwrote self-storage,
senior housing, and commercial real estate as a Strategic Investments intern at HPI Real Estate Services
and Investments, 30+ properties representing over $200 million in value.
I also co-founded a real estate wholesaling business, and when I couldn't find good enough data to run it,
I taught myself to build the systems myself. Today three businesses run on software I wrote, on schedules,
unattended, with zero paid data vendors.
Eight production systems, each built to solve a problem I had, each still running today, plus one in progress.
Statewide Foreclosure Lead Engine
★ LEGENDARY · 2026Active
A Texas-wide property-distress data platform built to replace paid lead vendors. County-specific
scraping engines plus PDF-extraction pipelines pull live records from 206 Texas counties, running
unattended every hour from 5 AM to 9 PM with zero paid data vendors. The delinquent-tax board alone
holds 228,239 leads across 56 counties, the buyer database has grown to 128,887 investors mined from
208,926 deed records, and leads are enriched with GIS spatial-join matching at a 98.6% match rate.
In October the whole fleet moved from a Windows laptop to a Mac Studio, where 42 scheduled jobs now
run it unattended.
Leaflet map dashboards on Cloudflare Pages are used daily by a business partner and two contract
call operators.
Demo: the statewide map and the delinquent-tax board (owner names and addresses hidden)
Outbound Calling and Texting Stack
◆ EPIC · 2026Live
Runs the outreach side of the wholesaling business for two remote callers. Leads flow into a Podio
CRM and a smrtPhone dialer, every call outcome and dialer note syncs back to the lead record on its
own, and a daily scorecard tracks each caller's time on the dialer. An automated texting engine sent
2,766 single-segment texts in its first three weeks, with hard guardrails: a stop, wrong-number, or
do-not-call reply is never texted again, and any other reply becomes a draft that waits for a human
yes. Alerts fire only when something fails, so silence means it worked.
Podio API
smrtPhone API
Node.js
PowerShell 7
macOS launchd
Screenshot coming soon
Demo video coming soon
Partner War Room and Comps Engine
◆ EPIC · 2026Live
The page my business partner and I work deals from. A map and call queue ranks every lead by what
needs attention first: a missed call, an overdue follow-up, a callback the AI flagged as owed, then
how warm the seller is. One save writes the status, offer, pass reason, and follow-up date back to
the CRM. A built-in comps engine prices each property from free public data only (sold comps, county
appraisal parcels, rents, and our own buyer file): after-repair value from median price per square
foot with a confidence grade, offer tiers at 65, 70, and 75 percent net of repairs, rental yield,
and who has been buying nearby. Every lead is archived before it leaves the CRM, 3,466 so far.
Python
Leaflet
Podio API
Cloudflare Access
> private by design: partner-only, behind sign-in
Demo: the Comp Desk deep dive, ARV, offer tiers and who buys nearby
AI Pulse: Daily Intelligence Engine
● RARE · 2026Active
An autonomous daily briefing system that reads 36 configured sources (RSS, Reddit, Hacker News,
arXiv), holds 1,993 stories in a rolling 21-day window, and hands only the genuinely new ones to a
headless LLM agent that ranks them into an HTML briefing every morning at 6:30 AM. The expensive
agent stage is deliberately optional: on the days it fails, the briefing still ships from the raw
ranked feed instead of not shipping at all.
Demo: one morning briefing, ranked stories and company filters (private notes hidden)
Private-Seller Vehicle Finder
● RARE · 2026Live
Built for a used-car dealership: scans Craigslist and OfferUp within a 30-mile radius,
values each vehicle with an LLM, decodes VINs through the free federal NHTSA vPIC API, and auto-flags
scam or bait listings before a human ever looks at them.
PowerShell 7
Claude API
NHTSA vPIC API
Screenshot coming soon
Demo video coming soon
AI Voice Interview Agent
◆ EPIC · 2026Live
A bilingual voice agent that conducts structured business-discovery interviews out loud: it works
through ranked question banks, generates domain-aware follow-ups from an industry knowledge base,
and knows when to offer a break or let a sensitive question go. 19 interviewers are deployed across
two client engagements, carrying 658 questions between them in English or Spanish. When a session
ends it processes itself: transcript, LLM analysis, a written report, and a follow-up list of
declined or thin answers routed to a human. The interaction design follows published HCI research
on interviewing older speakers: patient turn-taking, breaks offered but never imposed, and an
opt-out before every sensitive topic.
ElevenLabs Agents
Claude Opus
Python
Vanilla JS
> no public twin for this one: the code runs live client interviews and stays private
Demo: the interviewee app, language and voice choices (no live call)
AI Consulting Delivery Platform
● RARE · 2026Active
The system behind a consulting practice that brings AI into established businesses, currently a
used-vehicle dealership and a decades-old commercial landscaping operation. An 11-module bilingual
demo platform lets each owner see AI working on their own operation, an ideas portfolio holds only
proposals that survived research against the business's real constraints, and ranked interview kits
(362 and 296 questions) drive the discovery work the voice agent runs. Onboarding a new client in
any industry is three files and one deploy. For the landscaping client, a records app now turns
photographed receipts, invoices, and delivery tickets into verified project books: OCR, then an AI
reading step, then human sign-off. It read 241 of 243 key fields correctly in testing and is backed
by 361 automated tests.
HTML / CSS / JS
Claude Code
Python
FastAPI + React
PaddleOCR
> client work: no public code, no client names
Automated Hiring Funnel
● RARE · 2026Live
A four-round hiring pipeline that screened 265 applicants for a remote calling team without a
recruiter. It scrapes the inbox on a schedule, drafts every reply, ranks applicants on a written
rubric that rebuilds live as new answers arrive, and sends scheduled reminders through a journal
that makes double-sending impossible. A color-coded Excel workbook is the single source of truth.
The funnel produced 175 second-round responses, 38 finalists, and 13 structured interviews, run
end to end by one person and an agent.
Python + openpyxl
Gmail automation
Claude Code
Excel
JARVIS: Voice Operations Assistant
● RARE · 2026In Progress
A voice assistant for running the business hands-free. Say "Hey Jarvis," ask a question, and it
answers out loud: how the callers are doing today, whether the overnight jobs ran, how the county
scrapers are holding up, and which new skip-traced leads came in. Where it is now: v1 was
built in November 2026 with a wake word, local speech-to-text, a Claude brain that runs on my
existing subscription with no per-call API bill, and five read-only briefing tools, all tested in
text mode. Where it is headed: the live microphone loop tested end to end, a spoken morning
briefing on a schedule, a natural-sounding voice, read-only calendar access, and a menu-bar icon,
so I can check on the whole operation without touching a keyboard.
openWakeWord
faster-whisper
Claude Code (headless)
Python
> in progress: v1 built November 2026, text mode tested, live voice loop next
// how_i_build
My actual workflow
Not a philosophy, just what I've learned running eight production systems mostly solo.
01 Plan in plain language first
Before I write a line of code or a prompt, I write out the plan the way I'd explain it to my business
partner: no jargon, no assumed architecture. If I can't describe what a system should do in plain
sentences, I'm not ready to build it. That plan becomes the spec I hand to the model, and it's the
first thing I check when something breaks: usually the bug is that the plan was wrong, not the code.
02 Delegate the research, save judgment for what matters
I run most of my systems through Claude Code, and I've learned to treat model choice like a budget.
Cheap, fast models do the research: scanning county websites, summarizing scraped pages, checking
whether a scraper's output looks right. I reserve the expensive, high-reasoning model for the calls
that actually require judgment: architecture decisions, data-quality tradeoffs, anything where a wrong
call is costly to unwind. That split is what let one person build and maintain 27 county-specific
engines without burning unlimited time or money.
03 JSON is the source of truth, generated files are disposable
Every pipeline I run keeps its real state in structured data (JSON, a spreadsheet, a database), never
in a one-off script's local variables. Scripts, dashboards, and reports get regenerated from that data
whenever they're stale; I never hand-edit an output file, because the next run would just overwrite it
anyway. That discipline is what makes it safe to delete and rebuild almost anything in these systems
without losing real information.
04 Build for the person who isn't technical
My business partner and the operators who use these tools day to day aren't developers, and they
shouldn't have to think like one. The call pipeline is written only in Python's standard library, on
purpose, so a routine dependency update can't break a tool a non-technical person depends on to do
their job. The lead dashboards are a map with taps and clicks, not a database console. If a tool needs
a manual to use, I consider that a bug.
05 Automate for the failure, not just the success
Anything that runs unattended will eventually fail unattended: a site changes its HTML, an API
rate-limits, a scheduled task silently doesn't fire. I design for that from the start: runs that are
safe to retry, logs that say what actually happened, and scheduled jobs that pick back up rather than
requiring a person to notice and restart them. That's the difference between a demo and a system three
people depend on every day.
// campaign_log
One continuous story
I haven't missed a summer of work since high school, and two of my first employers are now clients of
systems I build. The tools just got better.
MSN-01May – Aug 2019
Lawn Maintenance Crew, Hollywood Garden Inc.
First job, landscaping in the South Texas heat: planting, nurturing, and learning what a day of labor is worth. The operations I saw there are now on my build list: I'm designing automations for this exact company.
MSN-02May – Oct 2020
Salesman, A1pha_Kicks
Sold vintage clothing and sneakers: managed stock, restocking, and order deliveries. First taste of retail and the resale game I'd later run myself.
MSN-03May – Aug 2021
Intern, Escamilla Used Cars & Truck Center
Shadowed the sales and finance managers at the family lot and helped facilitate 40+ vehicle sales. Years later I came back and built their private-seller acquisition tool.
MSN-04Jun – Jul 2022
Intern, Law Office of Mario Castillo PLLC
Shadowed attorneys across criminal, corporate, and personal injury law. A first look at the contracts behind every deal, years before I started negotiating my own.
MSN-05Jun – Nov 2022
Founder, Octavio's Curbside Painting
Started a residential painting business at the door-knock level: $25K in revenue, 150+ clients, my first two employees hired and trained, and a 1,000+ follower page doing the marketing.
MSN-062023 – May 2024
Founder, Imperial Kicks
Sold $20,000+ of sneaker inventory through supplier relationships across Texas, Los Angeles, and Florida, powered by a 2,000+ follower storefront I grew myself.
MSN-07Oct 2023 – May 2025
Leasing Agent, Housing Scout
Closed 50+ apartment leases and grew a 1,000+ follower real estate channel that generated its own lead flow.
MSN-08Jun – Jul 2024
Financial Analyst Intern, Kraus Real Estate Development
Built financial models for a $300M warehouse development spanning nearly 1,000 acres, created REIT comparisons to guide strategy, and shadowed a $40M loan pitch.
MSN-09Jun – Jul 2025
Real Estate Intern, GHG Investment Properties
Learned the wholesaling game from the inside: analyzed distressed properties, ran 1,000+ cold calls a week, and closed a $200K+ wholesale deal in San Antonio.
MSN-102026 – present
Real estate wholesaling, co-founder
Came back and restarted it as my own company this time: co-founded a statewide operation, hired two call operators, and built the data systems all of it runs on.
MSN-11Jun – Jul 2026
Strategic Investments Intern, HPI Real Estate Services and Investments
Underwrote self-storage, senior housing, and commercial real estate acquisitions in Austin: 30+ properties representing over $200 million in value, modeled with institutional-grade Excel builds. Built major additions to the team's internal asset management and acquisitions platform, still in daily use.
MSN-122026 – present
AI systems builder
Taught myself to build production AI systems with Claude Code: eight systems in production, 228,239 leads generated, three businesses running on software I wrote.
Why I'm making this move
I've been chasing the feeling of building something from nothing since I was in high school, when I
started a curb painting business and then a content page just to watch an idea turn real. Finance
taught me discipline, but this summer my internship introduced me to AI tooling, and building with it
gave me that same feeling every single day. I didn't just study the tools. I went home and built
systems that now run three businesses. This work sits exactly where I've always wanted to live:
creativity, business, and technology in the same place, moving faster every month. I want to spend my
career building at that intersection, at a company pushing AI forward, with the energy of someone who
found this a little late and refuses to waste a minute of it.
// skill_tree
Skill Tree
Levels are earned, not claimed. Every rating below traces back to something I actually built or logged.
County engines + PDF extraction across 206 Texas counties.
claude_codeLVL 9
Daily driver: multi-agent, headless, production schedules.
automation_opsLVL 8
42 scheduled jobs running unattended on one Mac.
financial_modelingLVL 8
CRE underwriting at a Strategic Investments desk.
sqlLeveling upLVL 3
Training arc in progress: MIS 325 + nightly reps.
españolLVL 10
Native.
// dispatches
Transmission log
Short updates as things ship. Newest first.
DSP-003November 2026
Moving day, then the deal desk
In October the whole operation moved from a Windows laptop to a Mac Studio, and 42 scheduled jobs
now run unattended there. On top of it shipped a partner War Room that ranks every lead by what
needs attention next, a comps engine that prices any property from free public data, a texting
engine that sent 2,766 texts in three weeks with hard do-not-contact guardrails, and a records app
for a consulting client that reads paper receipts into verified project books, backed by 361 tests.
DSP-002August 2026
The consulting era opens
Three new systems shipped in three weeks: a bilingual voice agent that conducts client discovery
interviews (19 deployed), an 11-module demo platform that shows an owner AI working inside their own
business, and a hiring funnel that screened 265 applicants down to 13 interviews. A video ad engine
also turned single product photos into three finished commercial spots. Meanwhile the data platform
kept compounding: 207 counties now feed the engine and the delinquent-tax board holds 247,088 leads.
DSP-001July 2026
Base established
octaviobenavides.com is live: four production systems documented, six public repos shipped, and two
one-page resumes wired to real PDFs. Next transmissions: demo videos for every build, the SQL training
arc, and application season.
// resume
Two versions, tailored to the role
Pick whichever matches the role you're evaluating me for.