Projects

Lenders and microfinance institutions often rely on fragile spreadsheets or generic CRMs that lack loan amortization models, automated collection channels, and double-entry accounting records, leading to balance mismatches and high delinquency rates.

Architected a specialized credit and debt-collection CRM with an automated French amortization engine, an immutable double-entry ledger, custom risk scoring, and WhatsApp API integration for automated collection alerts.

A ledger-first architecture was chosen over a generic CRM database model to guarantee that every single transaction is audit-trackable and mathematically immutable, rather than trying to patch accounting consistency onto a CRM designed for sales pipelines.

Powering the management of over RD$10,000,000 in active credit portfolios in the Dominican Republic, reducing average portfolio delinquency (morosidad) by 18% within the first 90 days.

Independent digital media platforms require rapid, real-time publishing workflows, SEO optimization, and high availability under sudden traffic spikes from trending news without complex infrastructure overhead.

Built a high-performance, SEO-optimized digital newspaper platform using Next.js, featuring dynamic section routing, server-side rendering (SSR), and automated image optimization.

Opted for Next.js Server-Side Rendering (SSR) over Static Site Generation (SSG) to guarantee that breaking news sections update in real-time for visitors, while serving fully pre-rendered SEO content for search crawlers without database latency.

Achieved sub-second page loads and optimized Core Web Vitals, ensuring zero downtime and seamless content delivery during breaking news events.

Every Rails AI project was reinventing prompt management differently, leading to fragmented, unversioned, and untested prompts in production.

Built a centralized prompt management engine utilizing Rails conventions that adds schema validation, versioning, localization, caching, and reusable prompt templates.

A conventions-first engine was designed so prompt updates could be version-controlled and tested in isolation like database schemas, rather than treating them as unverified strings or making API round-trips for configuration.

Streamlined prompt updates from code releases, cutting feature iteration cycles by 60% and enabling product teams to safely modify prompts directly in production.

Modern Ruby applications rely on unverified gem dependencies, exposing production services to supply chain vulnerabilities and malicious packages.

Designed an automated CLI tool that generates CycloneDX SBOMs (Software Bill of Materials) and audits gems for security vulnerabilities on every CI/CD run.

Designed as a zero-dependency CLI that generates standardized CycloneDX SBOM files in-memory to ensure audit speeds and security checks are extremely fast, without adding additional risk or dependency chains to the target pipeline.

Protected production pipelines against compromised gem releases with zero developer overhead, blocking vulnerabilities before deployment.

Bootstrapping a fully typed, scalable GraphQL server with Prisma and type-safe schema resolvers is repetitive and error-prone across different backend codebases.

Created a production-ready, pre-configured template featuring Pothos GraphQL schema builder, Prisma ORM, Yoga server, and Jest test suite.

Opted for code-first GraphQL schemas with Pothos rather than schema-first designs to enforce TypeScript compilation-time checks, ensuring GraphQL resolver inputs and return types are always in sync with Prisma schemas.

Eliminated API bootstrapping boilerplate, allowing developers to spin up complete, type-safe API servers with CRUD resolvers in minutes instead of days.

Go developers often lack a simple yet flexible ORM that doesn't introduce heavy, complex abstractions which obscure database queries.

Designed a lightweight, highly intuitive data mapper ORM tailored specifically to Go's idiomatic persistence patterns.

Decided to write a pure data mapper ORM rather than an Active Record clone in Go to prevent entities from holding active database connections, preserving strict boundaries between domain objects and data access logic.

Reduced database query complexity and connection overhead, enabling cleaner persistence codebases with minimal performance penalty.