01
CSN Analytics
featuredManufacturing · ML
Steel defect detection and cutting optimization for manufacturing.
PythonPandasML
~/portifolio
try me · shift+tab
❯
▋
01
About
I build production AI systems that ship — agent loops over ~90 typed tools, Knowledge-Graph RAG, and the benchmarks that grade them.
Focused on LLM systems: tool calling, prompt caching and context management, RAG/GraphRAG, MCP, and agent evaluation. At OceanPact I build conversational agents and a Knowledge-Graph RAG platform; at Turing I design multi-agent benchmarks and the verifiers that grade them. Open-source contributor to cognee. Portuguese (native), English (fluent), Spanish (intermediate).
Location
Rio de Janeiro, Brazil
Education
Computer Engineering @ UFRJ
Status
Open to opportunities
02
Projects
02
GraphMan
UFRJ · Graph Theory
A Rust graph library that measures itself, plus a browser observatory. Adjacency list, bitset matrix and CSR representations; BFS/DFS, diameter, components and Dijkstra over swappable frontiers, benchmarked on graphs up to 4.8M vertices. Compiled to WebAssembly, with traversals animated on the GPU.
RustWebAssemblyWebGPUNext.js
03
Debait
UFRJ · AI chatbot
A debate room with an AI referee. Deb scores every argument as it lands with the math visible, flags fallacies by quoting the exact words, and fact-checks on the web when a debater calls @deb. Rooms are event-sourced and streamed live over SSE; the sides fight over a vgpu (WebGPU) pixel grid behind the chat.
Next.jsAI SDKGroqSSERedisvgpu
04
CV Agent
UFRJ · Artificial Intelligence
An AI agent that reviews résumés with progressively loaded skills: role matcher with web search, format checker, 0–100 scorer and an editor that proposes changes as an inline diff. The agent decides which skill to load, and its findings are highlighted directly on the CV.
Next.jsAI SDKGroqAgent skillsvgpu
02.1
Open source
Contributions and proposals to tools I use
cognee
Contributor · AI memory library
ephemeral-context
RFC · Claude Code
RFC + TypeScript reference implementation of cache-aware eviction of used context (skills, files, tool results) in agent loops. A/B test: −34% cache reads, −7.2% input cost.
03
Skills
LLM Systems
Tool calling
Prompt caching
Context management
Agent evaluation
MCP
RAG
GraphRAG
OpenAI
Claude API
Claude Agent SDK
Languages
Python
Rust
TypeScript
JavaScript
SQL
C
C++
Java
Frameworks & Tools
FastAPI
Django
React
Next.js
Node.js
WebAssembly
WebGPU
vgpu
LiteLLM
LangChain
LangGraph
Databases
PostgreSQL
pgvector
LanceDB
Qdrant
Neo4j
FalkorDB
Redis
04
Experience
2026 — Present
AI SwarmBench Engineer (Contract)
Turing
Design multi-agent benchmark tasks that evaluate coordination in LLM agent swarms. Write verifiers that automatically grade agent outputs, and audit agent trajectories for correctness and quality.
2026 — Present
AI Developer
OceanPact
Built the conversational AI agent for a geospatial operations platform: agent loop with parallel tool-calling over ~90 typed tools and real-time SSE streaming with resume-after-reload (Python, FastAPI, OpenAI SDK, Postgres). Built a prompt-regression benchmark and logging to measure agent behavior across prompt changes. Built a production Knowledge-Graph RAG platform that ingests PDF/DOCX, extracts entities and relationships via LLMs, and persists them to a knowledge graph + vector store — polyglot persistence (pgvector, Neo4j, Redis), LiteLLM gateway, per-user ACLs (React 19, FastAPI).
2024 — 2025
AI & Automation Python Developer
Mundivox
Built and maintained Python tools automating optical-network monitoring and incident handling, including log parsing and reporting with Pandas. Built a local RAG chatbot giving the support team access to internal documentation and procedures.
2023 — 2024
Software & Automation Researcher
CBPF
Developed sensor-based automation systems and algorithms for real-time data processing at the Brazilian Center for Physics Research.
05Let's work
Let's work
together
Open to new opportunities and collaborations.