Blog
Designing AI agents, the Japanese market, and what we learn deploying them. Written as we go.
Adapting Sierra AI's τ-bench evaluation, deterministic coding, and state management for Bookmi.
2026-01-18 · 12 min
Analyzing METI's AI strategy, Japanese enterprise Agent adoption cases, and language-specific challenges.
2026-01-15 · 10 min
Why is keigo difficult even for GPT-4? Explaining Bookmi's 3-layer keigo model and evaluation system.
2026-01-12 · 8 min
Refund hallucinations, inventory misrecognition, inappropriate promises—revealing how we prevent these in real-time.
2026-01-08 · 14 min
Technical details on pgvector + Hybrid Search implementation, chunking strategies, and real-time updates.
2026-01-05 · 11 min
Dynamic model selection by language/task/cost, fallback design, and cost optimization explained.
2025-12-28 · 9 min
Analyzing AI adoption reality and challenges at major chains like Skylark, Saizeriya, and Sushiro.
2025-12-20 · 10 min
Maximizing Anthropic/OpenAI Context Caching features. Revealing system prompt design tips.
2025-12-15 · 7 min
Staff shortages are becoming critical. Learn how to use AI to improve efficiency and run your restaurant with fewer people.
2025-12-10 · 6 min
Significantly reduced labor costs and food waste through shift optimization and AI ordering. See how Company A boosted sales.
2025-12-05 · 8 min
The listing fee is just a figure in the books. Once you count how many of those customers come back and how many came only once, the real cost per customer tends to look very different from what you assumed.
2026-07-24 · 6 min read
Getting your licence isn't the finish line. Temperature records, hand-washing facilities, ingredient labelling — these come up most often in periodic inspections, and usually because nobody knew they were required.
2026-07-18 · 7 min read
The real damage from a bad review isn't the rating — it's how the next reader sees you handle it. Three typical cases, how to think about each, and reply templates you can adapt directly.
2026-07-11 · 8 min read
