RemoteYeah
K
kazusa931209's photo
Kazusa Tsubota
From Japan 06:57 AM (GMT+09:00)
$60,000/yr
Active 5 days ago

Share this profile:

AI Platform Engineer

Platform Engineer
Available for hire
Years of experience
7+ years
Experience level
Senior
Available from
08 Nov 2026

I am an AI Platform Engineer with 7+ years of software engineering experience across production AI infrastructure, cloud platforms, and backend systems. I connect GPU-based LLM serving with reliable production applications.

At ScalyX.ai, I designed vLLM inference infrastructure on Kubernetes/EKS, built AI Gateway and multi-model routing across LLM, speech, and vision workloads, and migrated 30+ tenants onto a shared EKS platform. These efforts reduced average monthly infrastructure costs by approximately 17% and GPU-related infrastructure costs by 15%. I also built real-time Whisper STT pipelines and production services using Python/FastAPI, RAG, embeddings, pgvector, and LLM tool calling.

My core tools include AWS, Kubernetes/EKS, Terraform, Docker, Helm, CI/CD, Prometheus, Grafana, OpenTelemetry, Jaeger, Python, TypeScript, NestJS, PostgreSQL, and Redis Streams. I focus on inference latency and throughput, observability, reliability, security, tenant isolation, and cost optimization.

I am based in Japan and looking for fully remote AI platform, AI infrastructure, production AI, or backend engineering opportunities with teams that can hire internationally.

Skills

TypeScript React.js Next.js NestJS Node.js Redis Docker AWS Software as a Service REST Kubernetes Terraform Microservices Speech to text PostgreSQL EKS CI/CD Generative AI Large Language Models Helm Charts Python FastAPI Grafana Cloudwatch Prometheus

Languages

English Japanese

Employment History

AWS AI Infrastructure / AI Production Engineer at ScalyX.ai Current 2025 - Now
Architected and migrated a multi-tenant AI-powered retail platform from per-tenant EC2 to a shared Kubernetes platform on AWS EKS, consolidating workloads across 30+ tenants and enabling demand-based autoscaling while reducing average monthly infrastructure costs by approximately 17%. Designed and optimized vLLM-based GPU inference infrastructure on Kubernetes/EKS, evaluating model characteristics, GPU requirements, workload patterns, and serving configurations to balance inference latency, throughput, availability, and cost. Implemented an AI Gateway and multi-model routing layer across LLM, speech, and vision workloads, routing requests based on workload and resource requirements and reducing GPU-related infrastructure costs by 15% while maintaining target service performance. Extended the AI platform to support real-time speech workloads, integrating Whisper-based STT with streaming audio pipelines and AI Gateway routing, with partial/final transcription handling for downstream LLM workflows. Built production AI services and workflows using Python/FastAPI, PostgreSQL, Docker, embeddings, semantic search, pgvector, RAG, and LLM tool calling, connecting retrieval, generation, and controlled backend actions. Designed scalable Kubernetes workloads using ALB, health checks, HPA, cluster/node autoscaling, and Helm, while standardizing infrastructure deployment with Terraform, Docker, ECR, and CI/CD. Built end-to-end observability with Prometheus, Grafana, OpenTelemetry, Jaeger, and CloudWatch, reducing average incident investigation time from 50 minutes to 30 minutes (40% reduction). Applied AWS/Kubernetes security and tenant-isolation controls using IAM, VPC/private networking, Security Groups, Kubernetes RBAC, Secrets Manager, and tenant-level access controls; performed synthetic load testing that validated aggregate request throughput of up to 10K requests/sec under controlled load while evaluating latency, concurrency, and GPU utilization.
Infrastructure / Backend Engineer at AllWeb3 2022 - 2024
Consolidated six backend services from six dedicated EC2 instances into a shared t3.medium environment using NestJS monorepo and Turborepo, preserving logical service boundaries while reducing the production EC2 footprint by 83% (6→1) and monthly EC2 cost from ~$45 to ~$30 (33%) for Alpha/MVP traffic. Designed and operated AWS production infrastructure across EC2 and Aurora RDS, right-sizing compute and database capacity against observed workload patterns rather than maintaining independent per-service capacity. Established a modular TypeScript/NestJS backend architecture with PostgreSQL and shared domain/API boundaries, collaborating with frontend engineers on feature specifications, API contracts, and implementation requirements while structuring services for future containerization and Kubernetes/EKS adoption. Automated non-production infrastructure scheduling with AWS Lambda and EventBridge, reducing development-environment runtime by approximately 70% (~50 vs. ~168 hours/week per instance) while preserving availability during engineering hours.
Full-Stack Engineer at StoreHub 2018 - 2021
Built commerce and POS capabilities using Next.js, TypeScript, React, Node.js, REST APIs, and PostgreSQL, supporting transaction, catalog, inventory, customer, and multi-location merchant workflows across a cloud SaaS platform. Designed reliable distributed and asynchronous systems using idempotent APIs, Redis Streams consumer groups, background processing, and business-rule validation for e-commerce, POS, inventory synchronization, order processing, and notifications. Optimized high-traffic product and inventory services with tenant-scoped Redis caching and PostgreSQL query/index/schema optimization, achieving a 90%+ cache hit rate and processing 1,000+ notifications/sec with <150ms p95 delivery latency.

Education

Bachelor of Engineering in Computer and Intelligent Systems Engineering at University of Malaya 2014 - 2018
Report this profile
Browse more Platform Engineer profiles