Frequently asked questions

Three models: fixed-scope project contracts (master + individual contract), a monthly technical-advisor retainer, and spot technical consultation. Project work is deliverable-based; advisory is billed by hours/month. Quasi-mandate (準委任) contracts are also available.

Roughly 120–160 hours/month for project work, and 10–40 hours/month for advisory. It varies with my current pipeline, so please reach out to confirm.

For a zero-to-one project starting from requirements, a small MVP is roughly 1–2 months and a mid-scale SaaS is 3–6 months. Start dates are typically 2–8 weeks out depending on my current bookings.

Yes. I can sign on your template or provide mine, and I'm happy to sign an NDA before any commercial discussion begins.

I have master and individual contract templates. If you have your own format, I'll follow it. Clauses (governing law, IP, liability caps, warranty) can be adjusted to your needs.

Project work is a fixed, deliverable-based quote; the monthly advisory retainer is roughly ¥200,000–800,000/month (by hours); spot consultation is about ¥25,000/hour. See the /pricing page for details.

As a rule, project work is 50% upfront + 50% on delivery; long engagements are billed monthly; advisory is month-end close, paid by the end of the following month. Transfer fees are borne by the client.

Work is remote by default, so routine meeting travel doesn't apply. On-site requirements gathering or surveys are billed at cost with prior approval (unless included in a retainer).

Yes. After a 30–60 minute consult I provide a rough estimate, and only propose a paid requirements phase when more precision is needed.

Frontend: TypeScript, React, Next.js (App Router), Vite, TanStack Query, Zod. Backend: Python (Flask, FastAPI, SQLAlchemy), Go (Echo, google/wire), Node.js. Infrastructure: AWS (Cognito, Lambda, ECS/Fargate, RDS, Bedrock, SES, CloudFront), GCP, Terraform, Vercel. Databases: PostgreSQL, pgvector, DynamoDB, Supabase.

Yes. For the lumber-industry DX B2B SaaS (a METI Minister's Award winner), I single-handedly handled the whole process — requirements, high-level design, frontend, backend, AWS/Terraform infrastructure, CI/CD, and operations.

Yes. I've built production-grade RAG systems with AWS Bedrock (Claude, Titan), pgvector, Pinecone, and LangChain — taking them past PoC to production quality with 90%+ accuracy and ~1.5-second responses.

Yes. Depending on the legacy code's quality, I may first run a tech-debt assessment (3–5 business days), then recommend whether a full rebuild or partial refactor fits — with the trade-offs made explicit.

Yes. On a restaurant-matching site for the tourism industry I joined GitHub-based team development (pull requests, code review, CI/CD). I'll follow your existing workflow (Scrum, Kanban, etc.).

Aligned with the OWASP Top 10: input validation with Zod/Marshmallow, authn/authz with AWS Cognito, asset protection via signed URLs, vulnerability testing with OWASP ZAP, and dependency scanning with pip-audit/npm audit in CI/CD. For B2B SaaS, I've met financial-institution-grade security requirements.

As a rule, business-logic and API layers get 70%+ coverage via unit + integration tests. For the frontend I recommend E2E tests (Playwright) over critical paths. For an early MVP, spec tests may take priority over raw coverage.

CloudWatch (or GCP Monitoring), error tracking with Sentry, and structured logging come standard. Advisory contracts don't include on-call standby, but I provide runbooks and self-recovery mechanism design.

As a rule, copyright and patent rights to delivered work are assigned to the client. Generic, reusable library/template parts may retain open-source elements, stated explicitly in the individual contract.

It's fine. Generative AI is an accelerator for implementation, test generation, and refactoring — I own the design decisions and ultimate responsibility for quality. AI-written code still passes human verification gates: TypeScript/Python type checking, unit tests, API/E2E tests (Playwright), and dependency/vulnerability scanning (OWASP Top 10). In practice: 2,153 automated tests on one backend, 96.7% type coverage on another, and zero double charges in production on a payments platform.

Pair-building with generative AI speeds up implementation, test code, and refactoring, and the saved hours go straight back into the rate. With one point of contact from design to infrastructure, there's no inter-vendor coordination cost, and serverless (Lambda, DynamoDB, Vercel) plus Terraform IaC minimize operating and maintenance costs. It's not cheap-and-poor — it's optimizing the cost structure without lowering quality.

Copyright to deliverables is assigned to you as a rule (stated in the individual contract). The AI tools I use are configured/contracted so input data is not used for training, and an NDA can be signed before any commercial discussion. I strictly avoid putting secrets or credentials in prompts, and manage secrets via environment variables and a secrets manager.

Yes. Beyond unit tests (Vitest / pytest), I design, implement, and run API integration and browser E2E tests (Playwright) myself. Tests are wired into CI/CD (GitHub Actions) to catch regressions automatically. Generative AI makes the test scaffolding that's often skipped for cost reasons affordable to include as standard.

See the /process page for details. In short, seven steps: initial consult → requirements → estimate & contract → design → implementation → QA → delivery & handover, with demos every two-week sprint.

Chat (Slack, Microsoft Teams, Chatwork), meetings (Zoom, Google Meet), and issue trackers (GitHub/GitLab/Backlog). I adapt to your existing workflow.

If it can be done fully remotely, yes — anywhere in Japan or abroad. With at least 5 hours/week of overlapping working time (JST), I can collaborate with overseas teams across time zones.

Business communication is primarily in Japanese. English documentation, code comments, and technical terminology are no problem.