Luis Aviles · Ineffable Solutions
I spent seven years running operations. Then I started building the software those operations needed. Everything on this page is running in production or published in the open: AI phone agents, custom platforms, developer tools, and the hundred-plus sites underneath them. I design it and write it myself, end to end.
Where this comes from
Seven years in operations before Ineffable: Windstream, LTC Technologies, then running operations at a cleaning company. Managing cross-functional teams, writing the SOPs, owning the numbers somebody upstairs wanted explained on Monday.
That is where the software came from. I kept hitting the same wall: the tool that would have fixed the problem either did not exist, cost more than the problem, or needed us to change how we worked to suit it. So I started building. The first Owner Command Center was not a product idea. It was me, at a cleaning company, tired of five systems that did not talk.
It means I read a business the way an operator does, not the way a vendor does. I know what a shift handoff actually looks like, why the workaround exists, and which report someone quietly rebuilds by hand every week. That is usually where the real work is.
The first three belong to one system, Telestation. It shipped in about three months.
The work
Pick a category. Everything listed shipped, and is either running now or published in the open.
AI agents & automation
Production systems that hold state, run queues, and cause real problems downstream when they break.
Platform + production console
The platform and internal production console for a phone-recording company (SIP Voice sub-brand) delivering professional auto-attendant, IVR and on-hold recordings to 2,000+ business customers. The voices are AI-generated and the console runs the entire production and rollout workflow. A script goes in; a broadcast-ready, mixed recording comes out minutes later.
My own command center
Two apps in one system: an admin console to run the agency, and a client analytics portal pulling Google Ads, GA4, Search Console and lead data into a single view. An autonomous AI agent picks up queued jobs on a dedicated server and finishes them end to end, around the clock.
End-to-end Python pipeline
Finds local businesses with weak websites, scores them, auto-builds a full custom demo site for each one, and deploys it live, so I walk into the call with their new site already built. Runs on open-source models (Qwen, GLM, DeepSeek) to keep the per-prospect cost near zero.
Marketing & customer acquisition
The same question every owner asks. I answer it with three engines running at once, and I validate the market with live API data before anyone spends a dollar.
The whole method, walkable
A start-to-finish walk-through of the acquisition engine I build for a business: what gets researched, what gets launched, and what gets adjusted weekly. Built as a live page rather than a deck, so a prospect can see the actual reasoning instead of a list of services.
Paid · organic · conversion
Paid search through Google Ads, organic and map discovery through SEO and Google Business Profile, and conversion work on the site itself so the traffic turns into calls and forms. Run separately they leak. Run together each one makes the other two cheaper.
Seven-phase process
Market sizing, keyword validation, competitive mapping, SEO roadmap, campaign architecture, creative, then launch and optimize. Every keyword, CPC and market assumption is checked against live data from DataForSEO, the Census Bureau, FRED and the Google Ads API first. No template ad groups and no autopilot.
Published on GitHub
These are the agents I run on real engagements, not demos written to make a README look busy.
Research pipeline
Seven-dimension market research across 15+ live data APIs, covering sizing, competitors, segments, regulatory and geography. It ends in a strategy document you can act on instead of a pile of raw output.
Pre-build intelligence
A ten-dimension pipeline across patents, academic papers, competitors, repositories and benchmarks. Built to run before you write code, so what you build is informed by what already exists.
Quality gate
A seven-component gate for AI agent skills, built on the Knowledge Activation research framework. It checks whether a skill holds up inside a working system, not just on its own.
Custom software & integrations
When the off-the-shelf tool would have meant changing how the business works, I built the thing that fits how it already works.
Monitoring and security telemetry
Continuous visibility across a client’s systems in one operator view: uptime, service health, security events and traffic, measured against industry benchmark standards. Built so somebody senior is watching the whole estate without needing a security team on payroll.
Commercial cleaning
Job metrics, crew performance and operational reporting for a commercial cleaning company. These are the numbers the owner actually runs the business on, in one place instead of five spreadsheets and a memory.
Go command-line tool + MCP server
A Go CLI and MCP server for a client's SharePoint document library, with a local SQLite mirror and an Excel intelligence pipeline on top. Cookie authentication, so it needed no Azure admin access, which is the only reason it was possible at all.
Desktop tools
Both started as my own daily friction. They run on my machine every day, which is the only endorsement I trust.
Windows tray app
Select text anywhere (email, browser, Word), hit a hotkey, and it rewrites it on the spot: fix the grammar, make it professional, cut it down. The API key lives in Windows Credential Manager rather than a config file.
Private, local transcription
Meeting transcription with no bot joining the call. It captures the audio on the machine itself and returns a timestamped, speaker-labeled transcript. Detects English or Spanish on its own and works with Zoom, Teams or Meet.
Production websites
Restaurants, home improvement, finance, faith-based. Built in Astro: fast, accessible, and yours outright instead of leased from a template platform.
telestation.com
Studio-grade phone recordings for operators: auto attendants, IVR and hold messages. The marketing face of the platform in the AI section, built around the line "turning wait time into quality time."
certipro.pro
Patio covers, pergolas and home remodeling across the DFW metroplex. Local SEO, schema and conversion tracking wired in from the start rather than bolted on after launch.
alexpatios.netlify.app
Alumawood, Knotwood and Equinox louvered covers across the Inland Empire. The whole site runs on one primitive: a scroll-driven light clock, so shadow angle, surface warmth and sky all move together and every interior page shares the home page film’s time of day.
horizonstudioshn.com
Architectural visualization: photorealistic 3D renders, walkthrough animations and 360° virtual tours for architects, developers and homeowners. A portfolio site where the imagery had to carry the sell.
sipvoice-site.netlify.app
White-glove business telecom: tailored, set-it-and-forget-it service with support to match. The sibling brand to Telestation, positioned for a slower, more considered buying decision.
Studio lab piece
A scroll film for a North Texas pool and pergola builder: the backyard becomes a resort as you scroll. Built in the Studio lab to push the cinematic scroll engine further than a client budget usually allows.
Volume, deployed
Built and shipped across client work and prospect demos in the last six months alone. Every one carries schema, analytics and a real Core Web Vitals budget rather than a theme and a prayer.
Design and motion arm
The website and visual production side: brand systems, motion design, real-time 3D exhibits, scroll-film builds and content production. Separate surface, same hands.
Heads up: the Studio runs a real-time 3D gallery in the browser. It needs a reasonably powerful machine. On older laptops or phones it will run slowly.
Case studies
Problem, approach, and what actually changed, all on the main site.
A financial intelligence platform that ingests, classifies and reports on every dollar, automatically.
Read itA classifier that auto-categorizes 95%+ of transactions and learns from every correction.
Read itAn automated extraction and analytics pipeline: every job, every cleaner, every metric.
Read itA 47-step manual process reduced to one command: topic in, finished video out.
Read itA buyer-ready financial package with dual P&L views, automated from raw data to presentation.
Read itA tool that learns a supervisor's correction patterns and flags issues before work is submitted.
Read itHow I work
Architecture, implementation, security posture and the AI layer. Nobody sits between you and the person writing the code, so nothing gets lost being relayed.
NIST SSDF with a 2,734-test gate, OWASP ASVS, SOC 2-ready practices. The standards are the floor here, not the pitch.
Agents, vector databases and local open-source models pointed at actual operations, with a cost model that survives contact with production.
Maintenance is part of the work, not a separate invoice. Everything built is yours; walk away whenever and you keep all of it.
Next step
Happy to walk through the code, the architecture, or the running systems, whichever is more useful to you. Behind all of it I run my own AI engineering system: roughly 120 custom skills and agents I use to ship this work fast.
