Skip to content

Experience

Experience

Where I've worked, what I built at each place and with which tools.

Experience

  1. Dec 2025 – Present

    Current

    PHINX Lab
    · Remote

    Team Leader & Senior Full Stack Developer

    AI platform built on an orchestrator plus skills. Senior since joining; Team Leader since June 2026.

    • I lead a team of 3 developers: technical planning, code reviews and architecture decisions, while staying hands-on in backend and applied AI.
    • I'm the main author of two of the platform's services — a document analysis agent and an MCP server — where the hard problem isn't calling the model but making the system deterministic and operable.
    • I designed a two-level orchestration architecture: a central orchestrator that dispatches to 7 structured-extraction skills, each with its own sub-orchestrator.
    • The orchestrator is modeled as a state machine: each analysis moves from one step to the next, the same pattern I used for case files at the Entre Ríos Education Council.
    • I built a 5-stage multimodal pipeline: vision, OCR, text extraction from digital PDFs (the ones that don't need OCR), regulatory verification and response synthesis.
    • I developed the regulatory verification system end to end: modeling and loading the legal corpus, indexing and search, evaluating compliance rules and generating the results report.
    • I exposed internal services over MCP — retrieval, technical report generation and document validation — as tools LLMs can consume.
    • Output determinism: state and concurrency guards in the analysis pipeline, contract tests for the skills and regression tests with real fixtures, so a prompt change can't silently break the format.
    • I instrumented observability: structured logs for Grafana/Loki and a correlation ID that follows each request end to end, from document upload to agent execution.
    • I found and removed rules that contradicted each other across different skills in the system: a class of bug that doesn't show up until the agent gives a wrong answer.
    • I designed a cached indexing and selective context-loading strategy that significantly reduced LLM usage costs and lowered document retrieval latency.
    • I built specialized workers and consensus engines for structured processing and decision-making in AI workflows.
    • I integrated models via the Model Context Protocol (MCP) and tool orchestration toward external systems, with agent and subagent architectures.
    • I maintain and fix a binary-format parser written in Python and invoked from the TypeScript pipeline: cross-runtime interoperability within a single workflow.
    • Earlier (Dec 2025 – Apr 2026), on a school management platform for the education sector, I developed the notifications module end to end: a Node.js and TypeScript service, the integration with the mailing provider, and the React interface.
    • Node.js
    • TypeScript
    • Fastify
    • Express
    • PostgreSQL
    • TypeORM
    • Redis
    • Python
    • React
    • Docker
    • Grafana/Loki
    • AWS (S3, Bedrock)
    • MCP

  2. Jul 2023 – Dec 2025

    Entre Ríos Provincial Education Council (Consejo General de Educación)
    · Paraná, Entre Ríos, Argentina

    Full Stack Developer

    Public sector. The Council (CGE) is the governing body of the education system in the province of Entre Ríos, Argentina.

    • I developed the province-wide Family Allowance Certificate Upload and Validation System (Salario Familiar): 20,000 to 30,000 requests per annual cycle and close to 50,000 files, with full auditing and traceability. It cut the average processing time from 4 months to 5 weeks.
    • I built the file-upload security layer from scratch in plain PHP — type validation, sanitization and access control — with no external libraries, due to a technical constraint set by the agency.
    • I led the Case File, Procedure and Internal Memo Management Platform: I digitized the paper-based workflows of more than 10 departments — around 60 people — with tiered permissions, a per-department work queue, traceability and audit logs on every case file.
    • I modeled the case file state machine — putting files on hold, closing them and reopening them with a new report — with requirements gathered together with the Payroll department.
    • I wrote the manuals and directed the tutorial videos for the system, for the province's entire education community.
    • I optimized SQL Server queries and bulk case-file loading, and refactored legacy code with a focus on performance, security and maintainability.
    • PHP 7
    • SQL Server
    • JavaScript (jQuery, AJAX)
    • Bootstrap
    • HTML5
    • CSS3

Stack

Languages
  • TypeScript
  • JavaScript
  • PHP
  • SQL
  • Python
  • HTML5
  • CSS3

I use Python for interoperability and maintenance inside Node pipelines, not to build complete systems. My working language is TypeScript.

Backend and APIs
  • Node.js
  • Fastify
  • Express
  • NestJS
  • Laravel
  • REST API design
Generative AI
  • LLM apps
  • Agents and subagents
  • Multimodal pipelines (vision, OCR)
  • Two-level multi-agent orchestration
  • Model Context Protocol (MCP)
  • Prompt and skill engineering
  • Multi-provider integration behind a common interface
  • Evals and output determinism
  • Amazon Bedrock

I work against a production RAG system and know its architecture in depth, but I didn't build the retrieval service: my work is the agents and skills that consume it.

Frontend
  • React
  • Next.js
  • Redux
  • Tailwind CSS
  • Bootstrap

I use Next.js in this portfolio (App Router and Server Components); in commercial production I work with React.

Databases and ORMs
  • PostgreSQL
  • SQL Server
  • Redis
  • TypeORM
  • Prisma
DevOps
  • Docker
  • Nginx
  • Git
  • AWS (S3, Bedrock; serverless fundamentals)
Practices and testing
  • SOLID
  • Clean Code
  • OOP
  • Modular architecture
  • Performance optimization
  • Jest
  • Scrum