I build production AI systems end to end, from the model layer to the interface.
Senior Full-Stack & AI Engineer
Hi, I'm Kiyoshi Araki. I'm a senior full-stack & AI engineer with 10+ years of shipping production systems across SaaS, e-commerce, and generative AI. I work across the full stack of an AI product: LLM and agent workflows, retrieval pipelines, computer vision, and on-device inference, along with the APIs, data layers, and infrastructure that keep them reliable. I'm strongest when the requirements aren't perfectly defined. I can help shape the scope, make sensible trade-offs, and deliver clean, maintainable implementations. If you want a partner who moves quickly while still caring about quality, I'd love to connect.
Tokyo, Japan
Location
10+ years
Experience
10,000+ users reached
Reach
Capabilities
AI & Machine Learning
Front-end
Back-end
Data
Cloud & Delivery
Notable Projects
Nonsense
Nonsense is a language-learning app from WordScenes that teaches 13 languages through Hollywood movies, with real-time translation, contextual definitions, and personalized learning paths. I architected the React Native app and own the speech and on-device AI layer, from offline pronunciation feedback to the Japanese NLP pipeline behind adaptive lessons.
- Architected a hybrid speech-to-text layer that switches between on-device and cloud inference, keeping voice response latency under 300 ms
- Compressed Hugging Face Whisper by about 70% with ExecuTorch and INT8 static quantization to run fully offline on-device
- Built cloud transcription on WhisperX and Replicate with character-level timestamp alignment for word-by-word pronunciation scoring
- Designed a Japanese NLP pipeline pairing JMdict and ichiran with word-frequency data over 50,000+ entries to rank words and sequence lessons
- Owned CI/CD and AWS infrastructure, taking deploys from about 45 minutes to under 10 with zero downtime
Zebracat AI
Zebracat turns text prompts, scripts, audio, or articles into ready-to-post social videos. I led the front-end architecture for the launch and built the generative pipeline behind it, from LLM content workflows to computer-vision scene analysis and the rendering backend.
- Led front-end architecture through launch, with Stripe billing flows that lifted subscription conversion from about 18% to over 22%
- Integrated OpenAI and LangChain RAG workflows that cut creation time per project from about 45 minutes to under 20
- Built PyTorch computer vision pipelines for scene analysis and avatar generation, cutting manual editing from about 8 steps to fewer than 3
- Optimized FastAPI rendering pipelines with Redis caching, reducing processing latency by 35% and holding SLAs through a traffic surge without extra infrastructure
- Owned AWS and Docker deployment pipelines with automated staging and rollback at a weekly release cadence
Job Wizard
Job Wizard is an AI-powered Chrome extension that autofills applications across LinkedIn, Greenhouse, and Workday, plus a job tracker and ATS-optimized resume generation. I built the extension UI and the form-field detection engine behind the autofill, backed by a Django REST API.
- Engineered the form-field detection engine across 100+ distinct site structures, lifting autofill completion from about 70% to over 80%
- Shipped the extension to 10,000+ active installs
- Introduced Jest testing and coverage baselines that sharply reduced regressions reaching production
Ashley x Descend Auction Platform
Ashley x Descend is an online furniture auction marketplace from Ashley Canada where prices drop over time and shoppers can either place a bid or grab the item with "Buy Now". I worked on the storefront and core commerce flows, and built the auction logic that powers the live price drops.
- Built the auction engine powering live price drops and bid matching
- Worked on the Next.js storefront where shoppers bid on items or grab them with Buy Now
- Implemented catalog, pricing, checkout, and delivery rules on Medusa.js
Menaji
Menaji is a U.S. e-commerce brand for men's skincare and cosmetics. I worked across the stack, from the customer-facing storefront to the API and data layer behind it.
- Worked across the React storefront, the Node.js API, and the MongoDB data layer behind it
- Integrated Shopify for product and catalog management
- Implemented a secure payment flow through checkout
Work Experience & Education
WordScenes
Senior Full-Stack & AI Engineer
Remote · U.S.
Architected the React Native app and own its speech and on-device AI stack, from offline Whisper inference and a hybrid on-device/cloud STT layer to the Japanese NLP pipeline behind adaptive lessons, plus CI/CD and AWS infrastructure.
Zebracat
Senior Full-Stack & AI Engineer
Remote · Germany
Led front-end architecture and Stripe billing for the platform launch, built the LLM, RAG, and computer-vision pipelines behind automated video generation, and owned deployment on AWS and Docker.
Job Wizard
Full-Stack Engineer
Remote · China
Built and maintained an AI-powered Chrome extension in TypeScript, including the form-field detection engine, its Django REST backend, and Jest test coverage.
FOURDIGIT Inc.
Full-Stack Developer
Minato City, Tokyo
Moved from front-end into full-stack ownership, pairing React and TypeScript interfaces with NestJS APIs, GraphQL, and PostgreSQL across client products.
Mahana Corporation
Front-End Developer
Shibuya City, Tokyo
Built responsive, accessible React and JavaScript interfaces for e-commerce platforms, and helped set the team's front-end review and performance standards.
Tokyo Institute of Technology (now Institute of Science Tokyo)
Bachelor of Computer Science
Meguro City, Tokyo
Contact
Best way to reach me is through: kiyoshiaraki.dev@gmail.com