Featured case studies
The strongest examples are shown first: the problem, what was built, and the evidence available today.
AI / RAG
RAG Knowledge Base
Problem: Document knowledge is hard to use when people cannot quickly find the right passage or verify where an answer came from.
Built: A public PDF knowledge assistant with ingestion, chunking, vector retrieval, source citations and refusal behavior for unsupported questions.
Evidence: Public demo. Independently tested by an AI-company founder, who specifically highlighted citation discipline and the decision to decline questions the source could not answer.
Next.js / Gemini / Supabase pgvector / PDF parsing
Multimodal AI / Workflow
Field Report AI
Problem: Field work can end with a second job: rebuilding photos, voice notes and observations into a client-facing report.
Built: A workflow that turns field inputs into structured information, keeps a human review gate, and generates a report PDF.
Evidence: Build, launch and distribution completed across Product Hunt, Contra and Upwork Portfolio. Third-party usage and paid revenue are not claimed yet.
JavaScript / Cloudflare Pages & Functions / Gemini / PDF
Localization QA / Chrome Extension
LocaleStress
Problem: Longer translations and right-to-left layouts can break a real interface before a team has full localization test coverage.
Built: A Chrome extension that stress-tests the current page with text expansion and RTL, highlights overflow and viewport issues, and exports a JSON report.
Evidence: Public MVP, public GitHub repository and downloadable release. Local-first design keeps page content and reports in the active browser instead of sending them to a server.
Chrome Extension / JavaScript / Local-first QA
Internal Automation / Orchestration
Web Agency OS
Problem: Moving one prospect through lead research, site audit, demo, sales and QA required repeated handoffs and created copy/paste errors.
Built: A thin orchestration layer over five independent tools, connected by one shared Lead ID and one lead workspace.
Evidence: Implemented internal system. Presented as workflow-architecture evidence, not as an external customer deployment.
Next.js / TypeScript / SQLite / HTTP integrations / CLI QA
CRM / Operations
Sales / Client OS
Problem: A small sales operation can lose opportunities because a follow-up, deadline or next action is forgotten.
Built: A local-first CRM and delivery OS covering lead → outreach → deal → client → production → maintenance, with mandatory next actions and funnel metrics.
Evidence: Automated tests cover pipeline logic, next actions, metrics, filters, exports, validation, lead import and lead-to-client conversion.
Next.js / TypeScript / SQLite / Zod / Vitest
Web / AI / API
KAISEI GLOBAL Web + AI/DX
Problem: A service business needs more than a brochure: multilingual content, inquiry paths, FAQs and integrations have to work together.
Built: Public service sites with English/Japanese content, inquiry flows, FAQ, AI guidance and backend/API integrations.
Evidence: Publicly deployed and used as KAISEI GLOBAL's own operating and sales infrastructure.
Next.js / React / TypeScript / Cloudflare / Supabase / Gemini / Resend / Drizzle
Paid client work
Alongside self-initiated products, this section shows real client delivery that reached completion and payment, with client-sensitive details removed.
Paid client engagement / Fiverr
Japanese Localization / Functional QA
Scope: End-to-end QA of a Japanese learner platform covering localization, UI copy, Japanese-input validation, content presentation and functional behavior.
Delivered: Delivered a structured final QA report with 10 evidence-based issues and verified completion across six learning flows.
Evidence: Real paid engagement completed in August 2026; order completion and settled payment were verified on Fiverr. Client identity, internal URLs and test-user information remain private.
Independent product feedback
An AI-company founder independently tried the RAG demo and highlighted two implementation choices: answers stay tied to cited passages, and the assistant declines questions the source material cannot support. The same reviewer described Field Report AI as a practical way to reduce the reporting chore that follows on-site work.
This is independent feedback on public prototypes, not a paid-client testimonial or customer deployment claim.
More public work
SaaS / Web App
Testimonial Collector
Submission → moderation → approved testimonial embed.
Web / Catalog
KAISEI Luxury Catalog
A responsive visual catalog prototype for high-value assets.
Game / Digital Product
Demon Gate
A 30-stage browser action game with a public demo and packaged buyer materials.
Game Dev / Localization
JP Ready for Godot 4
A public developer tool for Japanese-language and localization workflows in Godot 4.
Start with a small, testable scope
Have a manual workflow, AI feature, internal tool or web application that needs to be built? Share the current process, the painful step and the result you want. A defined paid pilot is welcome before a larger engagement.
Small AI / Next.js Paid Trial
Start with one real, bounded task
Typical range: US$150–250 / 2–3 business days
Good fits include one React / Next.js UI fix, one small API or AI integration, a bounded RAG improvement, or one workflow-automation step.
Final scope, fixed price and timing are confirmed after the brief. Delivery includes implementation, QA and concise handoff notes.