André Vinícius
AI Engineer & Software Engineer building Multi-Agent Systems, Machine Learning models, and scalable distributed backends.
About Me
I am a Software Engineering Intern at the Brazilian Federal Court of Accounts (TCU), an Undergraduate AI & Recommender Systems Researcher at UEFS, and a Google Student Ambassador 2026. Specializing in Python, TypeScript/Next.js, Go, and Elixir, I build production-grade Multi-Agent LLM architectures (Claude Agent SDK, MCP, deterministic guardrails, τ-bench evaluation), approximate nearest-neighbor recommendation engines (LSH Forest & MinHash), and fault-tolerant distributed backends.

Featured Engineering Projects
See all projects →Comanda — Multi-Agent Delivery Customer Support
Claude Agent SDK · Model Context Protocol (MCP) · FastAPI · Next.js · SQLite
Multi-agent orchestration (Reception on Claude Haiku + Order & Refund Specialists on Sonnet) with per-agent MCP servers, deterministic refund calculation in code, 4 tool guardrails, custom Portuguese BM25 + Snowball stemmer from scratch, and a 30-scenario τ-bench evaluation suite (8 adversarial, LLM-as-a-Judge, pass³) achieving 0 policy violations at $0.026/session.
LSH Forest & MinHash Clinical Recommender
Python · Datasketch · LSH Forest (l=200, k=20) · Serverless LRU Cache · MongoDB
Production User-Based Collaborative Filtering engine built at UEFS. Compresses patient histories into 200-permutation MinHash signatures indexed in self-tuning LSH Forest prefix trees, with a Memory-First LRU cache (~200ms warm start), batch inference (300 patients/call), deterministic clinical gender filtering, and 4-level graceful degradation.
Flow — High-Throughput Telemetry Pipeline
Elixir/OTP · Phoenix · Broadway · Oban · PostgreSQL
Asynchronous event ingestion and processing pipeline using Elixir/OTP and Broadway with configurable batching and backpressure, persistent background jobs via Oban, real-time Phoenix PubSub broadcasting, and Telemetry Metrics instrumentation.
Professional Experience

Tribunal de Contas da União (TCU)
June 2026 – PresentSoftware Engineering Intern
Applied Research & Engineering

Universidade Estadual de Feira de Santana (UEFS)
Undergraduate Researcher — Software Engineering & Recommender Systems
September 2025 – Presentl = 200 permutations, k = 20 neighbors) with a Memory-First Serverless LRU cache (~200ms warm-start latency).Leadership & Extracurricular Activities
Google Student Ambassador 2026
Awards & Recognitions

Instituto Claro / LSITEC
National Finalist — 13th Edition of Campus Mobile
Dec 2024 – Feb 2025Courses and Certifications
Platform Engineering (FAST)
80 hoursCESAR School

13th Edition Campus Mobile
Immersive WeekInstituto Claro / LSITEC
Recent Publications
Deep dives into Multi-Agent LLM orchestration, Model Context Protocol (MCP), Recommender Systems in production, and Neural Networks from scratch.
Como construir agentes de atendimento que não alucinam dinheiro: regras em código, guardrails MCP e avaliação τ-bench em português
Arquitetura do Comanda: orquestração multi-agente sobre Claude Agent SDK e MCP com cálculo determinístico de reembolsos em código, guardrails antes das ferramentas e avaliação automatizada estilo τ-bench com 30 cenários.
Sistemas de Recomendação em Produção com LSH Forest, MinHash e Cache LRU Serverless (latência de 200ms)
Como projetei e coloquei em produção na UEFS um motor de recomendação por Filtragem Colaborativa com LSH Forest (l=200), MinHash, Cache LRU em memória e degradação graciosa em 4 níveis.
Implementando Redes Neurais e Inteligência Computacional do zero em Python (Perceptron, Adaline, MLP, Kohonen, Fuzzy e Algoritmos Genéticos)
Como implementar a matemática fundamental de Redes Neurais Artificiais (Perceptron, Adaline LMS, MLP Backpropagation, Mapas de Kohonen) e Inteligência Computacional sem usar frameworks caixa-preta.
Core Tech Stack
Let's build something together
Feel free to reach out if you're looking for an AI / Software Engineer, have a question, or want to connect.