Looks like you have JavaScript disabled. For the full Twine experience, you will need to re-enable it.

Machine Learning Engineer focused on end-to-end AI system design, scalable deployment, and performance/cost optimization. I connect speech-to-speech pipelines with LLM reasoning, build adapter-based evaluation frameworks, and turn experiments into reproducible infrastructure. I also build and optimize AI for real-time applications—super-resolution/frame generation, autonomous driving planners, and game/client-server systems—while leveraging strong engineering practices in C++/Python, cloud deployments, and data/model compression.…Machine Learning Engineer focused on end-to-end AI system design, scalable deployment, and performance/cost optimization. I connect speech-to-speech pipelines with LLM reasoning, build adapter-based evaluation frameworks, and turn experiments into reproducible infrastructure. I also build and optimize AI for real-time applications—super-resolution/frame generation, autonomous driving planners, and game/client-server systems—while leveraging strong engineering practices in C++/Python, cloud deployments, and data/model compression.

Kangming Wu

AI Engineer, Data Scientist, Software Engineer, +1





Machine Learning Engineer focused on end-to-end AI system design, scalable deployment, and performance/cost optimization. I connect speech-to-speech pipelines with LLM reasoning, build adapter-based evaluation frameworks, and turn experiments into reproducible infrastructure. I also build and optimize AI for real-time applications—super-resolution/frame generation, autonomous driving planners, and game/client-server systems—while leveraging strong engineering practices in C++/Python, cloud deployments, and data/model compression.…Machine Learning Engineer focused on end-to-end AI system design, scalable deployment, and performance/cost optimization. I connect speech-to-speech pipelines with LLM reasoning, build adapter-based evaluation frameworks, and turn experiments into reproducible infrastructure. I also build and optimize AI for real-time applications—super-resolution/frame generation, autonomous driving planners, and game/client-server systems—while leveraging strong engineering practices in C++/Python, cloud deployments, and data/model compression.

Available to hire

Machine Learning Engineer focused on end-to-end AI system design, scalable deployment, and performance/cost optimization. I connect speech-to-speech pipelines with LLM reasoning, build adapter-based evaluation frameworks, and turn experiments into reproducible infrastructure.

I also build and optimize AI for real-time applications—super-resolution/frame generation, autonomous driving planners, and game/client-server systems—while leveraging strong engineering practices in C++/Python, cloud deployments, and data/model compression.

Skills

Experience Level

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Intermediate

Intermediate

Language

Chinese

Fluent

Japanese

Fluent

English

Advanced

Work Experience

Machine Learning Engineer at Fixstars Corporation

September 1, 2024 - Present

Built an end-to-end pipeline connecting a front-end speech-to-speech model to a backend LLM for asynchronous deep reasoning (modeled after GPT-Live-style architecture). Extended ServiceNow EVA with custom STT/TTS/LLM adapters to enable reproducible evaluation of Japanese phone-call agents under identical dialogue scenarios. Modeled per-call costs across STT/TTS/LLM combinations and selected a configuration that reduced projected cost by 12% vs baseline. Offloaded GPU-intensive STT/TTS/LLM inference from a monolithic server to AI Station by redesigning cloud–AI Station interfaces. Codified cloud and AI Station environments as Infrastructure-as-Code using Ansible and Docker for reproducible large-scale deployment. Also: ported PyTorch models to proprietary GPU by reimplementing neural network layers as compute shaders; refactored PLUTO data processing from Python to C++ to reduce processing time by 85%, reconstructed datasets to reduce training time by 50% and loss by 15%, and perform

Programmer at KOEI TECMO GAMES CO., LTD.

April 1, 2020 - August 31, 2024

Collaborated with product managers, artists, and external teams to design, code, debug, test, and develop new game features (equipment, events, characters, etc.). Fixed defects including logic errors, memory leaks, crashes, and fragmentation, eliminating 95% of reported errors. Nobunaga’s Ambition: Hadou (mobile game): designed schemas and client/server protocols; reduced Redis access by 95% and Redis keys by 90% for map unit launch data by moving logic into a zone manager class; monitored DAU/ARPPU and user activity metrics; set up and managed GitLab CI/CD machines for data conversion and compilation. Wild Hearts (co-developed with EA): implemented ECS-architecture programming; reduced average load time by 30% and reduced memory usage of UI text by 10% via object pooling; mentored new graduate programmers enabling code contributions within two months.

Machine Learning Intern at ALBERT Inc.

September 1, 2018 - December 31, 2018

Developed a ceramic-quality inspector to detect black dots on ceramic boards using deep learning.