AI Development Experience in Japan
Executive Summary
Senior Software Engineer with hands-on experience building production AI systems, LLM workflows, and AI-augmented engineering solutions across automotive and enterprise domains.
Focus Areas
- Production AI Systems
- LLM Integration
- AI Workflow Engineering
- Real-Time AI Applications
- AI-Augmented Software Engineering
1. Toyota AI Voice Sales Assistant
Period: 2022.11 - 2023.01
Designed a real-time AI assistant for sales conversations. The system analyzed customer speech during the call and surfaced key signals such as:
- budget
- family situation
- target vehicle
- financing intent
System Architecture
- Frontend: PowerApps / Flutter
- API gateway: WebSocket
- Backend: Lambda (Python)
- Speech processing: Azure Speech SDK
- Morphological analysis: MeCab
- Text analysis: Azure Text Analytics
- Storage: RDS / S3
- Output: live indicators on the frontend
Key Technologies
- AWS Lambda
- Azure Speech SDK
- Azure Text Analytics
- Python
- WebSocket
- MeCab
- RDS
Engineering Challenges
- Real-time speech streaming
- Speaker separation
- Timestamp alignment across services
- Lambda lifecycle recovery
- Multi-service orchestration
Technical Highlights
- Combined speech recognition and language analysis in a streaming workflow
- Used a hybrid cloud setup with Azure Speech for transcription and AWS Transcribe for speaker diarization
- Aligned results from multiple speech services using timestamp-based fusion
- Normalized timestamp differences between 100ns ticks and second-based outputs
- Distinguished customer speech from salesperson speech
- Designed checkpoint-based recovery for unstable Lambda lifecycles
- Decoupled high-frequency audio streaming from lower-frequency business logic
- Reassembled WebM audio chunks inside Lambda and converted them into continuous streams for recognition
- Defined a message protocol for multi-Lambda coordination
- Stored aggregated results in RDS
Business Result
The sales team could see live indicators on the tablet while speaking with the customer. This made the assistant useful as a real-time sales copilot.
2. AI Workflow Refactoring for n8n + LLM
Period: 2025 Q3
I refactored a prediction workflow for better maintainability, observability, and cost control.
Technical Work
- Split large workflows into smaller stages
- Reduced workflow execution time from hours to minutes
- Lowered token consumption and operating cost
- Improved observability and maintainability
- Moved complex business logic into backend services for better scalability
- Added automation around repeatable steps
Architecture Notes
- Replaced serial, one-record-at-a-time processing with staged batch execution
- Moved payload thinning and validation into the backend layer before model calls
- Used workflow nodes for orchestration, not for heavy business logic
Technical Outcome
The implementation reduced end-to-end execution time from roughly 3 hours to about 8 minutes and lowered operating cost to around $0.13 per run. The workflow also became easier to operate, debug, and extend.
3. LLM-Driven Engineering for Large Systems
Period: 2025.08 - 2026.01
I used LLM tools to accelerate understanding and delivery across backend, web, network, and cloud systems.
Main Activities
- Accelerated understanding of unfamiliar codebases
- Decomposed large tasks into executable engineering steps
- Troubleshot across backend, frontend, and infrastructure boundaries
- Used LLM-based tools to speed up code comprehension and implementation
- Recovered system context faster during multi-project delivery
- Kept the AI role as an engineering multiplier, not a replacement for design judgment
Tooling
- Claude Code
- OpenAI models
- Kilo Code
- GLM
Practical Positioning
I worked as a senior backend engineer and used AI tools to accelerate implementation, explanation, and learning. In practice, this meant I could handle backend work and the necessary frontend or integration work when needed.
Summary
My AI experience focuses on production AI systems and AI-enabled software engineering.
The common pattern across these projects is:
- combining AI services, cloud platforms, and real-time systems
- integrating speech, text, workflow, and backend components
- solving reliability, latency, and maintainability problems
- delivering production-grade systems
- using AI as a force multiplier for software engineering
