Marco Chen

Machine Learning Engineer WAT.aiJune 2026 - Present

— Researching on World Model for soccer dynamics, building a Selective SSM + Imagined Rollout + Actor-Critic architecture to predict next latent state and planning strategic actions.

Software Engineer Intern GeotabMay 2026 - Aug 2026

— Optimized harsh event recommendation latency, engineered an offline-first cache layer with a synchronization queue, built an error observability dashboard.

Game Engine Development Apprentice (AIGC) Tencent GamesJan 2026 - May 2026

— Designed a RAG persistent memory system and a hiearchical memory retrieval strategy.

Agentic AI Trainer ShipdDec 2025 - April 2026

— Reviewed and curated AI training datasets for LLM Coding Agents.

Software Engineer Intern GeotabSep 2025 - Dec 2025

— Built end-to-end compliance tests in C#, maintained and optimized the Geotab Drive App.

Software Engineer Intern Octopodi TechnologiesJan 2025 - April 2025

— Developed a coding plagiarism detection application in Rust and TypeScript.

LLM From Scratch

Implemented all components of a language model from scratch (tokenizer, model, optimizer), trained a minimal Language Model on TinyStory dataset.

Python, PyTorch, Transformers, CUDA, WandB, Multiprocecssing

Nano-vLLM

A minimalistic implementation of a transformer-based LLM inference engine optimized for low-latency and high-throughput on GPU with smart memory management.

Python, CUDA, Multi-Head Attention, KV Cache, Tensor Parallelism, Prefix Cache

LLM Gateway

High-performance API gateway for large language models with smart routing, semantic caching, rate-limiting, and real-time monitoring.

Go, gRPC, Redis, Kafka, PostgreSQL/pgvector, ClickHouse, Docker/Kubernetes, Prometheus/Grafana

High Concurrency Cache System

In-memory cache system optimized for high concurrency workloads with sharding and lock-free data structures.

C++, LRU/LFU/ARC

SnapNote

AI-powered note-taking platform that turns handwritten notes into organized digital documents with RAG semantic search and agentic workflows.

Python, TypeScript, OCR, RAG, LangGraph, PostgreSQL/pgvector, React

WLP4 Compiler

A compiler for the WLP4 language (Simplified C) that generates MIPS assembly code, implemented scanner, parser, semantic analyzer, and code generator from scratch using C++.

C++, Compiler, MIPS

Biquadris

A multiplayer Tetris video game implemented in C++. Supports both CLI and GUI, optimized resources management with RAII, implements software design patterns.

C++, RAII