AI ENGINEERING · SOFTWARE ENGINEERING · PRODUCT
I turn real problems into AI systems that reach production, using Software Engineering as the foundation, Product to define what matters, and AI where it truly delivers value.
Python, RAG, agents, LangGraph, OpenAI, PyTorch, Evals, FastAPI, PostgreSQL, pgvector, OpenTelemetry, TypeScript, Bun, Node.js, Docker, Terraform, and AWS.
Engineering with a product perspective.
01 / ENGINEERING METHOD
From business problem to production: domain, architecture, and constraints define where AI fits. Verification, security, and measurement make the system reliable and able to evolve.
Each case shows how far this method was exercised, with explicit evidence and limitations.
02 / SELECTED WORK
RAG and agents with execution controls, model evaluation and adaptation, rigorously evaluated Machine Learning, resilient backend systems, and automation. Each case presents explicit evidence and limitations.
All projectsAI Engineering · RAG · Agents
Production-oriented AI operations copilot with Python/FastAPI, evidence-grounded RAG, hybrid retrieval and bounded LangGraph agents; GitLab actions require human approval.
Evidence-grounded retrieval, deterministic authorization, approval-bound actions, evals and AI observability; independent post-release OpenAI and GitLab sandbox integration evidence.
AI Engineering · Model evaluation
Model evaluation and domain adaptation with a deterministic baseline, local LLM, hybrid policy, frozen held-out and human review; Jev/Laya experiments stay outside the runtime.
System implementation, frozen evaluation, and Laya adaptation with PyTorch/CUDA and validation-based selection; Jev and Laya remain evaluation-only.
AI Engineering · Machine Learning
Defect-risk ranking to prioritize code review. In the historical raw RF evaluation, reviewing ~30% of modules captured 71.26% of known defects.
Evaluation without duplicate-vector contamination, comparisons at equal review capacity, calibration, uncertainty and frozen-artifact inference.
Software Engineering · Reliability · Cloud
Transactional API with idempotency and failure resilience, validated in a temporary AWS lab and a reproducible local Docker benchmark.
End-to-end IaC, temporary deployment, troubleshooting, functional validation, operational evidence, destroy, and resource audit.
Software Engineering · Backend
Backend foundations in TypeScript, Node.js, Express, PostgreSQL and Prisma, with Clean Architecture, domain modeling, TDD, JWT authentication and per-user isolation.
Authored a TDD-developed API with JWT authentication, per-user isolation, ownership-based authorization, data integrity, and production hardening.
Web development · Performance
Web development for the agency website and blogs, including customization, new features, and performance work.
Full-stack work on the agency site and two WordPress blogs, spanning theme customization, performance, and server deployment.
Software · Operations
Development of an internal request management solution integrated with GitLab, in the context of EVAG’s operational needs.
Building the internal support solution and its GitLab integration to give demands a traceable flow.
Automation · Process
Process automation for generating petition pages, connecting Google Drive and WordPress.
Designed and built the automation that turns Google Drive content into published WordPress pages.
03 / EXPERIENCE & CAPABILITIES
From Product Owner to Software and AI Engineering: website delivery, production systems, process automation, and agentic development. Since October 2026, I work as an AI Engineer at Agência Catus and a Software & AI Engineer at EVAG.
More about my backgroundOct 2026 — Present
Oct 2026 — Present
Jun 2026 — Sep 2026
Apr 2024 — Sep 2026
Apr 2023 — Mar 2024
04 / ARTICLES
AI Engineering · Machine Learning
In DefectRisk, more complex models delivered small gains. Group-aware evaluation, conflicting labels and calibration showed where to invest next.
LET’S TALK
Professional opportunities, partnerships, or a conversation about AI Engineering, software, and product.
contato@marcelotaparelli.com.brConnect on LinkedIn