
Hiromi Shikata
Verified Expert in Engineering
Full-stack Engineer and Developer
Tokyo, Japan
Toptal member since February 12, 2025
Hiromi is a full-stack engineer and manager with 19 years of experience building scalable systems. An expert in TypeScript, Go, Python, Rust, and Clean Architecture, she brings strong skills in DevOps (GitHub Actions, auto testing), AI (OpenAI, Claude, AutoGPT), and infrastructure (AWS, GCP, Docker, Terraform). Focused on creating dynamic UIs, she leverages React, shadcn, and Tailwind. Hiromi leads teams using Agile and Lean methodologies and leverages modern tech stacks for impactful results.
Portfolio
Experience
- MySQL - 18 years
- OpenAI API - 15 years
- RESTFul APIs - 10 years
- Swagger - 9 years
- TypeScript - 9 years
- Go - 8 years
- Clean Architecture - 8 years
- Trunk-Based Development - 5 years
Availability
Preferred Environment
Ubuntu, GitHub, Trunk-Based Development, Agile
The most amazing...
...accomplishment in my career has been contributing to the success of multiple startups that achieved successful exits.
Work Experience
Full-stack Engineer and Engineering Manager
Welfare Product Replacement for Japanese Market Startup
- Resolved the inflexibility and cumbersome nature of the system structure, which had become a challenge when we decided to pursue an opportunity discovered during a pivot fully.
- Rebuilt the back end, front end, and infrastructure, which had bloated to about eight times the size needed for required functionality, to optimize for the new product.
- Leveraged AWS Lambda to replace Kubernetes, addressing burdensome maintenance efforts and costs.
- Evaluated the existing system, which used Flutter, and decided to switch to ReactNative based on the team's skill sets and recruitment market research.
- Hired new members to contribute immediately as effective team members, implementing an automated onboarding process since existing members were busy with other projects.
- Grew the team through member referrals rather than direct recruitment. Over three years, 30+ contributors joined, with only two resigning—one for entrepreneurship, the other for family care. This retention rate reflects high team satisfaction.
- Added assistants to boost our engineering hiring process when recruitment became challenging and conventional methods stalled. This proved effective, bringing in four strong candidates from over thirty applicants and enhancing the team.
- Minimized non-implementation time to manage tight schedules, as there were not many technical challenges due to the nature of the product.
Full-stack Engineer and Engineering Manager
AI Product Implementation at Startup
- Progressed from R&D results to product development, advancing to full-scale implementation based on business requirements.
- Separated the system architecture into a data and process management system and an AI-focused system to avoid using type-unsafe Python in areas requiring programmatic control, ensuring more stable programming.
- Built a serverless architecture using AWS Lambda as the execution environment to handle the simultaneous processing of thousands of requests required by the product.
- Implemented custom queue execution to overcome scaling limitations when the number of Lambda functions executable from SQS reached its limit, even though Lambda's individual limits were not an issue.
- Evaluated various models regularly and switched to better-performing ones as LLM options increased, with monthly improvements in AI performance and cost reductions.
- Architected an automated, asynchronous management system to handle the increased overhead as successful recruitment grew the team to approximately 16 members.
- Implemented remote work with flexible hours and no fixed meetings, using goal tracking and performance metrics. Streamlined engineering management to 2.5 hours daily, aiming to reduce it to under one hour.
- Introduced a daily evaluation system that uncovered previously undetectable issues, enabling problem detection and resolution by the following day despite the fully flexible environment.
- Implemented accurate individual engineer performance evaluations, enabling quantitative assessment for rate adjustments (increases or decreases) and team composition decisions, resulting in an estimated doubling of team cost efficiency.
Full-stack Engineer and Engineering Manager
Startup in Japan
- Built autonomous systems for task execution. LLMs struggled with multistep tasks, forgetting steps or making mistakes beyond three or four steps. Optimized prompts and task structures to fix the issues, achieving over 90% success in testing.
- Enhanced LLM task execution stability by identifying common error patterns based on task characteristics and addressing them through task decomposition and examples.
- Modified the system to integrate with OpenAI and two other major AI providers' APIs. Implemented multi-agent frameworks for complex tasks with autonomous correction, resulting in more versatile and reliable solutions.
- Led a team of four to six members, initially struggling with task allocation. With minimal AI knowledge in the team, we conducted research and met tight deadlines. Implemented agile processes to enable remote collaboration across multiple time zones.
- Enhanced client and investor demos with intuitive AI process visualizations to mitigate uncanny behavior. Introduced configurable settings for explanatory displays to optimize performance.
- Tested computer control via the operating system, focusing on visual interpretation. Single LLMs struggled with visual tasks, and multimodal systems performed better but lacked precision. Due to low success rates, alternative methods were explored.
- Explored retail product and origin detection techniques from five to six years ago with field specialists. While browser operations remained the primary focus, development centered on web interface control due to instability in image-based guidance.
- Tested a Chrome OSS extension for 3-step tasks but halted progress due to the high costs of integrating it with existing solutions. Developed methods to reduce token usage while preserving core information and staying within prompt limits.
- Achieved a 60 – 70% success rate with source code generation for automated operations. Improved to nearly 100% by implementing AI-friendly wrapper functions with programmatic internal processing to resolve a specific type of task.
- Focused on testing new LLMs, frameworks, and products in a rapidly evolving landscape. Team development led to peak costs of $75 per hour, prompting an investigation into open-source LLMs and newer options like Grok for cost and speed optimization.
Experience
Streamlining Code
https://github.com/HiromiShikata/ast-to-entity-definitionsSkills
Libraries/APIs
GitHub API, OpenAI API, React, OpenAPI
Tools
GitHub, Amazon Simple Queue Service (SQS), OpenAI o1, Shadcn, Docker Compose, Slack, NPM
Languages
TypeScript, Go, Python 3
Paradigms
Agile, Clean Architecture
Platforms
Ubuntu, AWS Lambda, Docker, Kubernetes
Storage
Amazon DynamoDB, MySQL, PostgreSQL
Frameworks
AutoGen, SST, Jest, Swagger, Selenium, Flutter, React Native
Other
Remote Work, Flexible Work, Lean, GitHubProjects, Domain-driven Design (DDD), Trunk-Based Development, Llama 2, Chrome Extensions, GitHub Copilot Chat, Sonnet 3.5, OpenAI GPT-4 API, OpenAI, Anthropic, AWS RDS Aurora, GitHub Issues, Gemini, RESTFul APIs, Writing & Editing, GitHub Actions
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring