About Me

Hi, I'm Martin (Mingtian) Chen — a backend systems engineer focused on database infrastructure, recommendation systems, and ML systems. I spent nearly three years at ByteDance, where I was promoted from SDE I to SDE II while building database middleware managing more than 10 PB of aggregate data. I am pursuing an MS in Information Networking at Carnegie Mellon University, expected Dec 2026.

Most recently, I built TikTok retrieval infrastructure in Seattle. I rebuilt the execution layer for all 29 retrievers and redesigned candidate-feature processing, cutting P50/P90 response-handling and parsing latency by 57%/71% while maintaining 99.981% field-level parity.

I have a cat named Mint (薄荷). He mostly supervises my coding.

Education

Carnegie Mellon University

MS in Information Networking | Pittsburgh, PA | Aug 2025 – Present | Expected Dec 2026

Selected coursework: Distributed Systems, Computer Networks, Cloud Computing.

University College Dublin

BSc in Software Engineering, First Class Honours | Beijing-Dublin International College | Beijing, China | Sep 2018 – Jul 2022

Experience

Software Development Engineer Intern, Recommendation Systems
TikTok
May 2026 – Aug 2026 | Seattle, WA
Rebuilt the retrieval execution layer for all 29 retrievers across vector ANN, inverted-index, behavior-graph, and GPU retrieval; consolidated logic covering 90% of replayed production requests into a shared platform.
Software Development Engineer II, promoted from SDE I
ByteDance
Aug 2022 – Jun 2025 | Beijing & Hangzhou, China
Owned DDL, metadata, and access-control modules for a Vitess-like Go middleware supporting 30+ teams, 100+ databases, and 10+ PB of data. Automated 50% of critical sharding and scale-out scenarios, migrated 60+ databases totaling 200+ TB with zero incidents, and increased global GTID allocation throughput by 10×.
Software Development Engineer Intern
Meituan
Jun 2021 – Dec 2021
Built next-day ETL validation to catch data inconsistencies and cut P99.9 latency in Java/Spring financial middleware by 50%.

Projects

Multithreaded x86 Operating System
CMU 15-410
Designed a preemptive x86 kernel with a 2 ms round-robin scheduler, 25 system calls, interrupt and exception handling, fork/exec/wait, virtual memory, and a user-level threading runtime.
BusTub Query Execution and MVCC
CMU 15-645 · 2nd overall on the course leaderboard
Built batch scan, hash join, aggregation, and window executors; added optimizer rules, Bloom-filter probe rejection, prefetching, composite-index lookups, and serializable MVCC with online garbage collection.
Blackwell GEMM Kernel Optimization
CMU 15-642
Reached 1.14 PFLOP/s for 8192³ FP16 GEMM on NVIDIA B200 using asynchronous TMA, multi-stage pipelining, persistent tile scheduling, and warp specialization; implemented a dual-MMA-consumer kernel that passed all correctness and timing tests.

Skills

Languages

CC++GoJavaPythonSQL

ML Systems

CUDATVM/TIRX

Systems & Infrastructure

LinuxMySQLRedisKafkaProtobuf gRPCDockerKubernetesAWS