Sr. Backend & LLM Engineer (AI agent) at Miquido

Led the development of an AI-powered chatbot for customer support, integrating machine learning algorithms and NLP. Implemented prompt engineering including CoT, Zero-Shot, and Few-Shot strategies to ensure consistent outputs. Configured infrastructure for fast inference at scale, handling 200–300 requests per minute. Implemented security measures (CSP, authentication/authorization with Auth0 in Okta). Built a secure backend with input validation to prevent SQL injection using ORM. Built a robust data backend with PostgreSQL and MongoDB via Prisma; optimized data access. Integrated voice AI with a TTS-based chatbot and established a RAG-style architecture to provide contextual responses.

Remote Backend Developer at Aozora Bank

Contributed to the development of a microservice-based architecture. Improved website response speed by 30% through Akamai CDN implementation, edge caching, Brotli compression, and early hints. Designed and implemented a microservice backend using Nest.js, enhancing flexibility and scalability. Established a strong CI/CD pipeline with Docker, Terraform, and GitHub Actions, while integrating ESLint, Prettier, and Jest. Enhanced data performance with Redis caching, and built a flexible workflow by integrating third-party services via webhooks (e.g., Comply Cube, Stripe, Okta). Implemented robust server-side caching to reduce data retrieval times.

Jr Backend Developer at Senda Japan

Developed a water risk management system using Express.js backend. Designed the database structure for seamless cross-platform communication with MySQL. Implemented CDN services via jsDelivr and Amazon S3, and built data flow for cross‑platform tracking and dynamic remarketing using Google Analytics and DoubleClick Floodlight. Collaborated with senior engineers on RESTful API architecture and backend integration.

Senior Backend & LLM Engineer (AI agent) at Miquido

Developed an AI-powered chatbot to assist customers, integrating machine learning and NLP. Implemented prompt engineering including CoT, Zero-shot and Few-shot approaches, and configured infrastructure to support fast LLM inference with 200-300 requests per minute. Built a secure backend with CSP headers and Okta authentication, created a robust data stack using PostgreSQL, MongoDB, and Redis, and implemented a scalable job queue with BullMQ. Designed RAG architecture using llm-index and embedding models to provide context from historic data. Improved output consistency and reduced hallucinations. Integrated Parler-TTS for AI voice, and connected third-party services via webhooks to support real-time workflows.