Production AI — not prototypes

Ship AI features your users actually trust.

I'm Yovany — a senior engineer who's put production AI into an enterprise platform serving premium clients. I help SaaS teams turn AI prototypes into reliable, shipped product.

Book a free 20-min fit call See results

Free, 20 minutes, no sales pitch — just a straight read on your AI.

8+ years · Immunefi · Thrive Market · Remote from México

AI Product Engineer · Remote, México

Trusted with mission-critical platforms

ImmunefiThrive MarketLuxoftBridgeDocEncargogo

The problem

Your AI demo impressed everyone. Then it met real data.

Most AI features look great in a demo and fall apart in production — hallucinated answers, brittle pipelines, no observability, and edge cases nobody planned for. Shipping AI that's actually reliable takes senior engineering judgment, not just a clever prompt.

A prototype that works on 10 documents but not 10,000.

No visibility into why the model gives the answers it does.

Features stuck at "almost production-ready" for months.

Proof

Outcomes, not adjectives.

Real numbers from production systems I've built and owned. measured in prod, not slideware

99.5%

accuracy on an AI-driven migration of 400 programs

−35%

support response time via an AI support system

50%

reduction in Time-to-Interactive & Total Blocking Time

+40%

Core Web Vitals improvement, lifting conversion

AI agents · what I build now

Production AI agents end-to-end: a document-intelligence RAG agent at BridgeDoc, AI-assisted data agents at Encargogo, and WhatsApp AI agents now in build at Citania.

citania.mx

Services

Three ways to work together.

Concrete, scoped engagements — no hourly rates, no open-ended retainers.

01AI Feature SprintFrom prototype to production in ~4 weeks.

I build and own the feature end-to-end — RAG pipelines, agents, and AI workflows with evals, guardrails, and observability so they hold up with real users and real data.

Book a call

02Fractional Senior EngineerEnd-to-end ownership for early-stage teams.

Embedded senior/lead engineering: database, frontend, backend, and deployment — shipping features, not tickets.

Book a call

03AI Architecture ReviewA clear, prioritized path to production-grade.

I get into your codebase and AI pipeline, then deliver a written diagnostic: where reliability, cost, and failure modes will bite, and a prioritized, concrete plan to fix them — so you know exactly what to do next.

Fixed scope · written deliverable

Book a call

Selected work

Production systems, owned end-to-end.

Live in production

Sendero Home

IoT access control & surveillance for condominiums

senderohome.pro

3: condominiums live
200+: active residents
6: cameras streaming

Problem

Condominiums run security and operations on spreadsheets, WhatsApp, and legacy desktop software — no resident app, no real hardware integration, and clunky access control nobody trusts.

What I built

Designed and own the platform end-to-end: a multi-tenant Next.js + Hono dashboard and API (PostgreSQL/Prisma, Supabase auth), an Expo/React Native resident app, and the hardware layer — ESP32/ESPHome gate controllers driven in real time through AWS IoT Core (MQTT + device shadows) with per-device mutual-TLS provisioning. Live entrance video streams from on-site cameras via go2rtc (RTSP→HLS) over Cloudflare Tunnel, secured with HMAC-signed URLs validated by a Cloudflare Worker and played natively in the app.

Result

In production across 3 condominiums with 200+ residents opening gates from their phones daily, area-based access permissions, and live camera check-ins at the entrance.

Building now

Citania

AI agents for WhatsApp

citania.mx

Problem

Businesses live in WhatsApp, but their support, sales, and follow-up are still manual — slow replies, leads going cold, and no way to scale conversations.

What I'm building

Building Citania: conversational AI agents that work natively inside WhatsApp. They hold real conversations, answer from a business's own knowledge, qualify and route leads, and trigger backend workflows — with guardrails and human-handoff so they stay reliable on real customer traffic.

Outcome

A WhatsApp-native AI agent platform that turns the channel customers already use into an automated, always-on front line.

BridgeDoc

A document-intelligence AI agent for US construction firms

Problem

Teams buried in hundreds of project documents, with no fast way to find answers.

What I did

Owned the platform end-to-end (DB, frontend, deploy to Vercel + a custom VPS pipeline) and built a Retrieval-Augmented Generation (RAG) agent: a conversational assistant that retrieves across hundreds of documents and answers with grounded, cited responses — plus the chunking, embedding, and retrieval tuning that keep those answers accurate at scale.

Result

Users ask questions in plain language and get instant, source-grounded answers instead of manually searching hundreds of documents.

Encargogo

AI-assisted data agents + e-commerce stabilization

Problem

An e-commerce platform with reliability and search-performance issues, limited visibility into user behavior, and no scalable way to keep product data fresh.

What I did

Owned development and stabilization; introduced observability and analytics; optimized Meilisearch and re-tuned the MongoDB schema/indexing; and built AI-assisted scraping agents that aggregate and normalize product data across Amazon, eBay, and Shopify.

Result

A more reliable platform with faster search, optimized conversion flows, and product data kept current by automated agents instead of manual entry.

About

Senior engineering, end to end.

I'm Yovany Luis, a senior software engineer based in Jalisco, México, with 8+ years building and owning mission-critical platforms — from Web3 security infrastructure at Immunefi to high-traffic e-commerce at Thrive Market. I work independently with SaaS and startup teams who need someone who can own delivery end-to-end and ship AI that survives contact with production.

As a father of two, I value efficiency and quality in equal measure — every solution I build is meant to be scalable, reliable, and maintainable long after I'm gone.

LinkedIn ↗GitHub ↗X ↗Résumé ↗Email ↗

Questions

Before you book.

How is this different from hiring an AI freelancer off a marketplace?

Most “AI freelancers” are prompt jockeys — they wire up a demo and disappear. I'm a senior engineer who has shipped production AI inside an enterprise platform and owns full delivery: database, backend, frontend, and deploy. You get evals, guardrails, and observability, not a clever prompt that breaks on real data.

What does the free fit call actually cover?

20 minutes to understand your AI feature, where it's stuck, and whether I'm the right person to help. You'll leave with a straight read on your biggest reliability risk — whether or not we work together. No slides, no pressure.

Do you only do AI work?

AI is the focus, but I own delivery end-to-end. If shipping your AI feature means fixing the data layer, the pipeline, the frontend, or the deploy, that's included — I ship features, not tickets.

What size teams do you work with?

SaaS companies and funded startups — typically seed to Series B — that need production-grade AI shipped reliably without hiring a full in-house AI team. US-first, LATAM-friendly, fully remote from México.

How do you price engagements?

Fixed scope, not hourly. The Architecture Review is a flat fee with a money-back guarantee; sprints and fractional work are scoped on the fit call so you know exactly what you're paying for and what you'll get.

Let's talk

Let's make your AI features production-ready.

I take a small number of new clients each month so every engagement gets senior attention. If your AI needs to be reliable, let's talk.

Book a free 20-min fit call

Currently booking · limited engagements per month

Not ready to talk? Grab the free guide:

5 Reasons AI Features Fail in Production (and how to fix them)