Luís Guilherme

Senior Software Engineer | AI & Distributed Systems Architect Remote (Brazil Based) · lgpelin92@gmail.com · github.com/lguibr · luisguilher.me

PROFESSIONAL SUMMARY

Senior Software Engineer with over 10 years of experience, bridging the gap between theoretical complexity (Physics background) and production reliability. I specialize in the intersection of Agentic AI workflows, Distributed Actor Models, and High-Performance Full Stack Architecture. I design and deploy production-grade systems, ranging from multi-agent LLM orchestrations and low-latency high-frequency trading interfaces to custom 3D game engines and distributed reinforcement learning pipelines. My mathematical foundation enables me to solve complex computational problems, ensuring systems are both theoretically sound and commercially scalable.

TECHNICAL ARSENAL

  • Core Stack: TypeScript (Expert), Python (Advanced), Go (Systems),React, Node.js.
  • AI & Data: LLM Orchestration (LangChain/LangGraph), RAG Pipelines, MCTS, AlphaZero, GumbelMuZero, PyTorch, Vector Search, TensorFlow.js.
  • Architecture & Game Dev: Event-Driven Systems, Actor Models, Microservices, WebSockets, WebGL, React Three Fiber (R3F), Procedural Generation.
  • Infrastructure: AWS, GCP, Docker, Kubernetes, Ray (Distributed Compute), Terraform.

PROFESSIONAL EXPERIENCE

Trebu (YC S22) | AI Software Engineer Jul 2024 – Nov 2025 (1 year 5 months) Architected autonomous negotiation engines and AI workflows for a high-growth logistics startup.

  • Multi-Agent Orchestration: Designed and deployed the Find Me a Truck protocol, an autonomous multi-agent system that negotiates pricing with truckers via WhatsApp, replacing manual operations with deterministic state machines and LLMs.
  • Latency Optimization: Reduced LLM pipeline latency by 66% through parallel execution trees, semantic caching, and aggressive prompt engineering. Slashed end-to-end negotiation time from 3+ minutes to <60 seconds, directly increasing booking conversion rates.
  • Computer Vision: Integrated Segment Anything Model (SAM) into the OCR pipeline, improving document parsing accuracy by 6% for non-standard paperwork.

Paradigm | Software Engineer (High-Frequency Trading) Apr 2022 – Apr 2024 (2 years) Engineered mission-critical frontend systems for the world’s largest institutional crypto derivatives network.

  • Performance Engineering: Built high-frequency trading interfaces requiring sub-second updates, capable of handling millions in daily volume without rendering lag or frame drops.
  • System Unification: Consolidated three disparate codebases (React Web, React Native, Electron) into a single monorepo architecture, reducing feature delivery time.
  • Reliability: Implemented critical-path E2E testing (Playwright) for trade execution flows, preventing regressions in complex financial and state management logic.

Trio | Tech Lead May 2020 – Apr 2022 (1 year 11 months) Technical leadership for multiple squads, delivering enterprise traceability solutions.

  • Architecture & Scale: Led the migration from monolithic structures to scalable microservices on GCP/AWS using Terraform.
  • Leadership: Mentored 8+ developers, introducing strict code review standards and CI/CD pipelines that reduced deployment failures to near zero.

SELECTED ENGINEERING PROJECTS

Projects demonstrating Systems Thinking, AI, and Advanced Architecture.

Daicer (AI Tabletop RPG Engine) | TypeScript, LangGraph, React Three Fiber, Strapi

  • Architecture: Engineered a multiplayer tabletop RPG platform powered by an AI Dungeon Master, fusing deterministic dice mechanics with generative AI storytelling.
  • Execution: Built the backend using LangGraph for agentic orchestration and Strapi for state management. Developed the 3D cinematic client (daicer-ui) using React Three Fiber and WebGL for real-time rendering.

Pongo (Distributed Multiplayer Game Engine) | Go, WebSockets, React, WebGL

  • Architecture: A distributed, real-time 4-player hybrid of Pong and Breakout with a custom 3D web client.
  • Execution: Built a custom Actor Model library (Bollywood) from scratch in Go to handle high-concurrency state management without race conditions. Synchronized state to the frontend via zero-dependency, low-latency WebSockets.

AlphaTriangle (Game AI) | Python, Ray, C++, PyTorch

  • Architecture: An AlphaZero-style AI agent trained via distributed self-play for a custom hexagonal puzzle game.
  • Execution: Implemented Monte Carlo Tree Search (MCTS) in C++ for maximum execution speed, wrapped in Python via Pybind11. Orchestrated distributed training jobs across multiple cores using Ray.

Tricked (Game AI & Mathematical Engine) | Rust (PyO3), Python, PyTorch, Svelte 5

  • Architecture: A high-performance reinforcement learning ecosystem implementing Gumbel MuZero to solve a custom puzzle game.
  • Execution: Architected a zero-cost mathematical verification engine in Rust, bound to Python via PyO3, while providing sub-second live spectator telemetry via a Svelte 5 frontend.

MimeFlow (Computer Vision Engine) | TypeScript, TensorFlow.js, WebGL

  • Architecture: Browser-based pose estimation engine for interactive applications and gamified fitness.
  • Execution: Utilizes 5D vector cosine similarity to calculate movement accuracy against video feeds in real-time. Optimized the pipeline to run entirely on the client-side GPU via WebGL, ensuring zero server latency.

LuisGuilher.me (Interactive OS Portfolio) | Next.js, Virtual File System, Monaco

  • What it is: A portfolio that simulates a complete Operating System / IDE in the browser.
  • The Tech: Engineered a Virtual File System (VFS) and window manager, featuring a fully functional code editor and command-line interface.

EDUCATION

Universidade Federal de Minas Gerais (UFMG) Advanced Undergraduate Studies in Physics (2018–2022)

  • Focus: Computational Physics, Complex Systems, Linear Algebra, and Calculus.
  • Completed extensive coursework before pivoting to full-time Software Engineering to apply mathematical models directly to production systems.

CEFET-MG Technical Degree in Information Technology (2006–2008)

Luís Guilherme

Luís Guilherme

Senior Software Engineer | AI & Distributed Systems Architect Remote (Brazil Based) · lgpelin92@gmail.com · github.com/lguibr · luisguilher.me

PROFESSIONAL SUMMARY

Senior Software Engineer with over 10 years of experience, bridging the gap between theoretical complexity (Physics background) and production reliability. I specialize in the intersection of Agentic AI workflows, Distributed Actor Models, and High-Performance Full Stack Architecture. I design and deploy production-grade systems, ranging from multi-agent LLM orchestrations and low-latency high-frequency trading interfaces to custom 3D game engines and distributed reinforcement learning pipelines. My mathematical foundation enables me to solve complex computational problems, ensuring systems are both theoretically sound and commercially scalable.

TECHNICAL ARSENAL

  • Core Stack: TypeScript (Expert), Python (Advanced), Go (Systems),React, Node.js.
  • AI & Data: LLM Orchestration (LangChain/LangGraph), RAG Pipelines, MCTS, AlphaZero, GumbelMuZero, PyTorch, Vector Search, TensorFlow.js.
  • Architecture & Game Dev: Event-Driven Systems, Actor Models, Microservices, WebSockets, WebGL, React Three Fiber (R3F), Procedural Generation.
  • Infrastructure: AWS, GCP, Docker, Kubernetes, Ray (Distributed Compute), Terraform.

PROFESSIONAL EXPERIENCE

Trebu (YC S22) | AI Software Engineer Jul 2024 – Nov 2025 (1 year 5 months) Architected autonomous negotiation engines and AI workflows for a high-growth logistics startup.

  • Multi-Agent Orchestration: Designed and deployed the Find Me a Truck protocol, an autonomous multi-agent system that negotiates pricing with truckers via WhatsApp, replacing manual operations with deterministic state machines and LLMs.
  • Latency Optimization: Reduced LLM pipeline latency by 66% through parallel execution trees, semantic caching, and aggressive prompt engineering. Slashed end-to-end negotiation time from 3+ minutes to <60 seconds, directly increasing booking conversion rates.
  • Computer Vision: Integrated Segment Anything Model (SAM) into the OCR pipeline, improving document parsing accuracy by 6% for non-standard paperwork.

Paradigm | Software Engineer (High-Frequency Trading) Apr 2022 – Apr 2024 (2 years) Engineered mission-critical frontend systems for the world’s largest institutional crypto derivatives network.

  • Performance Engineering: Built high-frequency trading interfaces requiring sub-second updates, capable of handling millions in daily volume without rendering lag or frame drops.
  • System Unification: Consolidated three disparate codebases (React Web, React Native, Electron) into a single monorepo architecture, reducing feature delivery time.
  • Reliability: Implemented critical-path E2E testing (Playwright) for trade execution flows, preventing regressions in complex financial and state management logic.

Trio | Tech Lead May 2020 – Apr 2022 (1 year 11 months) Technical leadership for multiple squads, delivering enterprise traceability solutions.

  • Architecture & Scale: Led the migration from monolithic structures to scalable microservices on GCP/AWS using Terraform.
  • Leadership: Mentored 8+ developers, introducing strict code review standards and CI/CD pipelines that reduced deployment failures to near zero.

SELECTED ENGINEERING PROJECTS

Projects demonstrating Systems Thinking, AI, and Advanced Architecture.

Daicer (AI Tabletop RPG Engine) | TypeScript, LangGraph, React Three Fiber, Strapi

  • Architecture: Engineered a multiplayer tabletop RPG platform powered by an AI Dungeon Master, fusing deterministic dice mechanics with generative AI storytelling.
  • Execution: Built the backend using LangGraph for agentic orchestration and Strapi for state management. Developed the 3D cinematic client (daicer-ui) using React Three Fiber and WebGL for real-time rendering.

Pongo (Distributed Multiplayer Game Engine) | Go, WebSockets, React, WebGL

  • Architecture: A distributed, real-time 4-player hybrid of Pong and Breakout with a custom 3D web client.
  • Execution: Built a custom Actor Model library (Bollywood) from scratch in Go to handle high-concurrency state management without race conditions. Synchronized state to the frontend via zero-dependency, low-latency WebSockets.

AlphaTriangle (Game AI) | Python, Ray, C++, PyTorch

  • Architecture: An AlphaZero-style AI agent trained via distributed self-play for a custom hexagonal puzzle game.
  • Execution: Implemented Monte Carlo Tree Search (MCTS) in C++ for maximum execution speed, wrapped in Python via Pybind11. Orchestrated distributed training jobs across multiple cores using Ray.

Tricked (Game AI & Mathematical Engine) | Rust (PyO3), Python, PyTorch, Svelte 5

  • Architecture: A high-performance reinforcement learning ecosystem implementing Gumbel MuZero to solve a custom puzzle game.
  • Execution: Architected a zero-cost mathematical verification engine in Rust, bound to Python via PyO3, while providing sub-second live spectator telemetry via a Svelte 5 frontend.

MimeFlow (Computer Vision Engine) | TypeScript, TensorFlow.js, WebGL

  • Architecture: Browser-based pose estimation engine for interactive applications and gamified fitness.
  • Execution: Utilizes 5D vector cosine similarity to calculate movement accuracy against video feeds in real-time. Optimized the pipeline to run entirely on the client-side GPU via WebGL, ensuring zero server latency.

LuisGuilher.me (Interactive OS Portfolio) | Next.js, Virtual File System, Monaco

  • What it is: A portfolio that simulates a complete Operating System / IDE in the browser.
  • The Tech: Engineered a Virtual File System (VFS) and window manager, featuring a fully functional code editor and command-line interface.

EDUCATION

Universidade Federal de Minas Gerais (UFMG) Advanced Undergraduate Studies in Physics (2018–2022)

  • Focus: Computational Physics, Complex Systems, Linear Algebra, and Calculus.
  • Completed extensive coursework before pivoting to full-time Software Engineering to apply mathematical models directly to production systems.

CEFET-MG Technical Degree in Information Technology (2006–2008)

luisguilher.me - Visual Profile Code

|

Made with ❤️