# James Yamazaki

James Yamazaki is a remote worker on Himalayas, Machine Learning Engineer at Sumsub, based in Kazakhstan.

Source: https://himalayas.app/@jamesyamazaki

```text
Research engineer at Sumsub who built a C++ face-recognition inference stack ranked 29th in NIST FRTE.
```

| Detail | Value |
| --- | --- |
| Availability | Looking for a job |
| Location | Kazakhstan |
| Authorized to work in | Kazakhstan |
| Salary expectations | 125k+ USD |
| Job categories | Research Engineer, Machine Learning Engineer |
| Skills | Machine Learning Systems, Distributed Training, Inference Optimization, Large Language Models, Computer Vision, Face Recognition, Vector Search, MLOps, Model Evaluation, PyTorch, Jax, C++, Python, Zig, Onnx |

## About

```text
At Sumsub, I built the C++ inference stack for its NIST FRTE 1:1 submission, which ranked 29th of 444 entries on the Mugshot track. I also developed an end-to-end face-recognition pipeline and deployed ONNX inference and vector search over 100M+ embeddings.

At Zencoder.ai, I shipped fill-in-the-middle code completion for VS Code and JetBrains IDEs. I implemented repository-aware retrieval and built an evaluation harness for large legacy repositories.

At INUI Gaming, I owned the inference stack for image classifiers serving a product supporting 10k+ concurrent users. I converted PyTorch models to TensorRT and introduced dynamic batching and request coalescing, reducing cloud inference cost by 23%.

At GOAT.AI, I co-authored The Poison of Alignment and helped release the resulting 7B model weights on Hugging Face. I also built multi-node LLM training infrastructure, including a 64-H100 deployment. Earlier, at Cerebra and Halbestunde, I worked on medical-imaging models, MLOps, and production inference.
```

## Experience

### Machine Learning Engineer

[Sumsub](https://himalayas.app/companies/sumsub). Sep 2024 - May 2026

```text
Built the C++ inference stack for Sumsub's NIST FRTE 1:1 submission, ranked 29th of 444 evaluated entries on the Mugshot track. Deployed ONNX-based inference and vector search over 100M+ embeddings.
```

Skills: C++, Onnx, TensorRT, Face-Recognition, Vector-Search, EC2, MLOps

### Research Engineer

Zencoder.ai. Jan 2024 - Aug 2024

```text
Shipped fill-in-the-middle code completion for VS Code and JetBrains IDEs. Implemented repository-aware context retrieval using AST- and path-based indexing and Tree-sitter symbol extraction.
```

Skills: FIM, VSCode, JetBrains, AST, Tree-Sitter, Harness, Human-Eval, SWE-Bench

### Research Engineer (Inference)

INUI Gaming. Aug 2023 - Jan 2024

```text
Served image classifiers for a product supporting 10k+ concurrent users with HPA and Triton Inference Server. Converted PyTorch models to TensorRT with layer fusion and INT8 quantization, reducing cloud inference cost by 23%.
```

Skills: PyTorch, TensorRT, Triton-Inference-Server, HPA, INT8-Quantization, Dynamic-Request-Batching, Kubernetes

### Research Engineer

GOAT.AI. Jun 2023 - Aug 2023

```text
Co-authored The Poison of Alignment, showing that removing refusal-style responses from SFT data improved MMLU, BBH, HumanEval, and DROP performance by 4-33%. Built and operated multi-node training infrastructure for LLM experiments on 64 H100 GPUs.
```

Skills: LLM, SFT, DeepSpeed, NCCL, NVIDIA-H100, A100, Hugging-Face, MMLU, BBH, Human-Eval, Drop

### Research Engineer

Cerebra. Oct 2022 - Jun 2023

```text
Led research on unsupervised anomaly detection in 3D CT using diffusion models and GAN priors, improving Dice by 20% relative to baseline. Built and owned the research team's MLOps stack, reducing deployments from hours to minutes.
```

Skills: Diffusion-Models, Gan, Computed-Tomography-(CT), MLOps, Experiment-Tracking, CI-CD, Docker

### Research Engineer

Halbestunde. May 2021 - Sep 2022

```text
Built an object-detection system for musical symbols using MMDetection, improving mAP@50 by 0.20 absolute over baseline. Deployed a production-ready inference service on Amazon SageMaker, reducing cost by 60% versus previous EC2 deployment.
```

Skills: MMDetection, Object-Detection, DVC, AWS-Sagemaker, EC2, Python

## Education

### Nazarbayev University

Bachelor of Science, Computer Science. 2018 - 2022

```text
Pursued a Bachelor of Science in Computer Science, graduating in 2022.
```

## Tech stack

- Python
- C++
- PyTorch
- DeepSpeed
- Docker
- CUDA
- Amazon SageMaker
- Hugging Face
- Zig
- JAX
- AWS

To contact James, message them from this profile with a Himalayas employer account: [Hire on Himalayas](https://himalayas.app/hire).

## People also viewed

- [Arsen Abzhanov](https://himalayas.app/@arsenabzhanov): AI Engineer Intern, Home Credit Bank
- [Alimzhan Bekmurzin](https://himalayas.app/@alimzhanbekmurzin): Data Scientist / ML Engineer, Treeple AI
- [Kudaibergen Abutalip](https://himalayas.app/@kudaibergenabutalip): Founding AI Engineer, Aldente AI
- [Михаил Котюшев](https://himalayas.app/@0005118): Senior Research Engineer, GeoAlert
- [Maksim Miller](https://himalayas.app/@maksmiller): Senior Software Engineer, AI Inference, NVIDIA
- [Kuat Zharykbayev](https://himalayas.app/@kuatello): Machine Learning Engineer, SpaceLab LTD
- [Aidar Amangeldi](https://himalayas.app/@aidaramangeldi): AI Engineer, Medical Center of the President’s Affairs Administration of
- [Alidar Aset](https://himalayas.app/@alidaraset): Machine Learning Engineer, Victory Technology
- [Erdauit Torekhan](https://himalayas.app/@erdauittorekhan): Lead Computer Vision Researcher, Qoima Automation
- [Arlan Toktarkhan](https://himalayas.app/@arlantoktarkhan): Machine Learning Engineer, BTS Digital