I ship production systems alone, end to end, and prove the hard parts with tests. Backend in Java 21 / Spring Boot — double-entry payments cores, idempotent transfers, concurrency control — plus web front ends and production LLM agents. Programming Technician, UTN (Dec 2025), Mar del Plata, Argentina. The architecture and invariants are mine; AI accelerated the implementation, and I can defend every line.

🏆 Top 8 — Anthropic & Kaszek Buenos Aires Hackathon 2026 💼 Automation Developer @ Wortise (contractor, 100% remote) from October 2026 🌎 Fully remote (UTC-3, strong US-hours overlap) · open to remote LATAM / USD roles

🤖 ai-booking-receptionist — public, runnable The AI core of my shipped WhatsApp booking agent — the sanitized layer extracted from my production app (VINDA).

• Bounded agentic tool-use loop over the Anthropic Messages API — hand-written, not the SDK's tool_runner or an agent framework
• 6 JSON-schema tools behind an allow-list
• Two-layer prompt-injection guard (regex pre-filter + stored-data sanitizer), both layers unit-tested
• 7 deterministic eval cases + prompt-cache wiring

Python.

💸 LedgerMind — public · runs locally in one command A double-entry payments core in Java/Spring. Append-only ledger (money as integer cents), idempotent transfers, and optimistic-locking concurrency. Verified by a 50-concurrent-transfer money-conservation test on real PostgreSQL (Testcontainers). Ships with a read-only MCP audit interface secured by OAuth2.1 (an AI agent can read the ledger, never move money), a SHA-256 hash-chain for tamper-evidence, Docker, and CI on every push.

📅 VINDA — production SaaS (private repo · in maintenance since July 2026) A multi-tenant scheduling & payments platform I built and shipped solo, full-stack:

• Backend — Java 21 / Spring Boot, Spring Security + JWT, WebSocket (STOMP) real-time chat & notifications, MercadoPago + Google Calendar integrations, Hibernate row-level multi-tenancy.
• Production Claude agent — the AI receptionist above, built on the raw Anthropic Messages API over the JDK HttpClient (no SDK client), plus per-tenant monthly LLM cost caps, an automatic Sonnet→Haiku degradation once a tenant's daily token usage runs high (cost control), and a Resilience4j circuit breaker.
• Frontend — React 18 SPA (booking, credits, chat, onboarding) with defensive XSS hardening (DOMPurify + server-side jsoup allowlist).
• DevOps — Docker, nginx/Caddy, Prometheus + Grafana observability.

Across VINDA, the architecture and invariants are mine; AI accelerated the implementation.

(The public repos above are self-contained; VINDA stays private.)

Languages: Java · Python (the agent repo above) · JavaScript · SQL Backend: Spring Boot · Hibernate/JPA · REST · Spring Security / OAuth2.1 · Spring AI (MCP) · learning FastAPI Frontend: React · HTML · CSS Data & Infra: PostgreSQL · Flyway · Docker · GitHub Actions (CI) · Prometheus + Grafana · nginx/Caddy AI: Anthropic / Claude API (agentic tool-use loops, prompt-injection defense, evals) Testing: JUnit · Testcontainers · Playwright · Vitest / Jest Starting Oct 2026 at Wortise: Node.js · TypeScript Security posture: defensive — parameterized queries (JPA), XSS sanitization, prompt-injection guards, auth gates

📧 juanfranciscopaezz@gmail.com · 💼 LinkedIn · 🐙 GitHub · CV on request