Staff Applied AI & Platform Engineer
I build the AI platform layer — the agent tooling, evaluation systems and infrastructure that let engineering organizations ship LLM-powered software into production, at scale, under financial regulation.
Nine years building in the LATAM payments ecosystem, now leading platform, architecture and applied AI at Deuna, a fintech in the Banco Pichincha ecosystem.
In 10 seconds
Download CVOperating scale
Not a pilot and not a prototype. I designed and built an internal AI developer platform wired into the delivery lifecycle of 300+ microservices in a regulated payments environment.
A Jira story goes in; a reviewed PR comes out.
Four-plane architecture — experience, control, execution, knowledge — on EKS with a LiteLLM gateway handling routing, fallback and per-team cost control. A refinement cockpit turns a Jira story into an execution brief, dispatches it to a sandboxed agent, runs the tests and returns a PR with an architect in the loop.
Agent output that survives a risk committee.
Golden datasets, LLM-as-judge scoring and regression gates on agent output quality, backed by continuous multi-cloud benchmarking across cost, latency and correctness. The layer that makes agent output defensible to a risk committee.
Every diff, every day, reviewed by an agent.
Pull-request review agent running today against every diff as a quality gate inside Azure DevOps pipelines — 20+ PRs a day across 300+ microservices, with cost and model performance tracked per run.
Tests written, run and triaged by agents.
Five agents mapped across a 16-step functional flow and a 9-step performance flow: impact analysis, test-case and automation-code generation, regression triage, and K6 threshold calculation — with the regression gate controlling deployment.
One source of truth every agent reads from.
Postgres-backed retrieval layer with AI-generated per-microservice and product-level documentation, so every agent reasons from one source of truth about how services actually interact.
How a bank says yes to AI.
Founded Deuna's AI Center of Excellence and its intake framework, pioneered Claude on Bedrock in production, and led the joint Deuna–Pichincha evaluation of enterprise AI platforms (Copilot Studio, Gemini Enterprise, Claude).
Refinement cockpit · IDE · Jira surface
LiteLLM gateway · routing, fallback, cost per team
Sandboxed agents on EKS · tests · PR emission
pgvector retrieval · generated service documentation
Five agents mapped across a 16-step functional flow and a 9-step performance flow, with the regression gate controlling deployment.
Determines what a change actually touches before anything runs.
Derives the functional cases the change requires.
Writes the executable test code.
Classifies failures — real regression or flake.
Sets the performance thresholds each run is judged against.
Nine years of payments infrastructure, reduced to the things that were hard to build and are still running.
Designed a QR payment flow that completes without connectivity, so merchants and users can transact when the network is down — the normal condition across much of Ecuador.
A decisive capability for financial inclusion.
Designed and shipped the integration with Ecuador's Central Bank instant payment system on ISO 20022 messaging and EMVCo QR, alongside a server-driven UI re-architecture of the payment flows.
Deuna on the country's interoperable rails.
Facial biometrics and liveness detection integrated with national civil registry systems, sitting behind KYC for both merchant and consumer products.
Regulated identity, owned end to end.
Spearheaded the migration off Fargate onto Amazon EKS and built the containerization foundation underneath it.
Still runs Deuna's production workloads today.
Re-platformed identity for live merchants, redesigning authentication, the role model and session handling for enterprise SSO and conditional access.
Zero downtime, on a live payments processor.
Architected and built the merchant super-app: payments, sales management, cashier operations, permissions and reporting.
Adopted by 35,000+ businesses nationwide.
Moobile — Lead Developer: migrated the Candy Club platform from legacy PHP to Angular. · WebCreek — Developer: built Aegis Flow for energy-sector stock operations (Angular + .NET Core) and led the Trestles Ionic mobile app.
Stack
Education
Beyond work
Open to remote
Staff-level applied AI and platform work — agent tooling, evaluation systems, and the infrastructure underneath. Regulated environments welcome.