TIPS

Every step, written down.

How we build healthcare data apps with generative AI — run for real, with no steps skipped. More articles are available in Japanese.

Claude Certified Architect – FoundationsClaude Certified Architect – ProfessionalWritten by an Anthropic Claude Certified Architect (Foundations & Professional)

FEATURED SERIES

Turning Visual Checks into Automated Checks
— Claude Code × TLF Layout

TLF Visual QC automatically checks page breaks and figure colours in TLFs (tables, listings and figures) that run to hundreds of pages, without the PDF ever leaving the browser. This post covers what it does, where it helps, why it is designed never to receive your outputs, and an overview of the series.

  1. Part 0Overview and Design
  2. Part 1Detecting Page Breaks — Measuring White Space by Greyscale Intensity
  3. Part 2Verifying the Results — Pixel Matching and an Answer-Keyed Dummy TLF
  4. Part 3Checking Colours — Greyscale, Colour Difference and Colour Vision
  5. Part 4Publishing and Operation — A Design That Never Receives Your TLFs

FEATURED SERIES

From In-House BI to an AI Dashboard
— Claude Code × Japan's DPC Open Data

Can the BI dashboards that companies build in Tableau or Power BI, for sales trends, area maps and more, become AI dashboards you can simply talk to? We tried it on Japan's public DPC open data (FY2022–FY2024, 6,448 hospitals). This post covers what changes, what the screens look like, how the AI is kept from making up numbers, and an outline of how we built it with Claude Code.

  1. Part 0Overview and Design
  2. Part 1Preparing the Data — Reconciling Years, Hospitals and Regions
  3. Part 2Checking the Numbers — Keeping Them Correct with Checks and Reviews
  4. Part 3Adding Generative AI — Deciding What to Leave to the AI
  5. Part 4Publishing and Running It — Managing Cost and Security

FEATURED SERIES

Building Systematic Review × AI Subagents with Claude Code

A five-part series rebuilding TrialMind, an LLM paper on literature search, screening and data extraction, with Claude Code subagents. Part 0 covers the original paper and this trial's question, conditions and flow.

  1. Part 0The Original Paper and This Trial's Design
  2. Part 1Search Query Construction and Subagent Roles
  3. Part 2Screening and the Subagent/Hook Split
  4. Part 3Data Extraction and Full-Text Handoff
  5. Part 4Discussion and How to Reproduce

FEATURED SERIES

Building Medicare Part D × Text-to-SQL with Claude Code

A five-part series on building a medical-data app with Claude Code. The theme is agentic coding — what you hand the agent, where you stop it, how you verify it. Part 0 covers the big picture and why we chose US open data.

  1. Part 0Overview and Harness Design
  2. Part 1Environment Setup and MCP
  3. Part 2Project Documents and Skills
  4. Part 3Prompt Design and Hooks
  5. Part 4Evaluation and Operations

Articles6

日本語の記事一覧 →