3 days of thinking.
6 hours of building.

Background

A small elder care home with a real differentiator
and zero digital presence.

Aannapurnaa Aai is an elder care facility in Borivali West, Mumbai — run by a close friend. It serves elderly people dealing with Alzheimer’s, dementia, Parkinson’s, paralysis, stroke recovery, and other conditions requiring full-time care. The facility caps at 18 residents — not a limitation, a philosophy: stay small, care deeply.

I designed and built the complete website pro bono — strategy, information architecture, visual system, copy, code, content, and SEO. One person doing the work of a strategist, designer, frontend engineer, and content team, with Claude Code as the execution multiplier.

Impact

Three days of strategy and system design. Six hours of building. Page 1 on Google within a week. 45+ pages, 12 service pillars, 30+ cluster articles, multilingual (English / Hindi / Marathi), PageSpeed 97 / 94 / 100 / 100.

The brief vs the real problem

The brief said “fix the website.”
The real problem was positioning and discoverability.

The reframe

The 18-resident cap wasn’t a weakness.
It was the product.

Positioning followed: “A home, not a facility.” Everything downstream — visual direction, photography brief, content tone, hierarchy of trust signals — came from this one strategic call.

The system

4 Claude Code skills and a research-first content pipeline,
before a single line of production code.

Speed is a byproduct of clarity. The 6-hour build was only possible because 3 days of thinking had already answered every question Claude would otherwise have needed to resolve mid-execution.

I wrote 4 skills before any production code. Each one exists because I identified a failure pattern Claude would hit at scale, and hardcoded the fix into a SKILL.md file. They are quality control systems, not shortcuts.

Blog content was research-first. For every topic, Claude queried DataForSEO for ranking keywords, search volume, and competitor content — then wrote articles built for ranking from the first word. Not blogs that were tagged later. Blogs designed from the keyword data up.

The work

A home, not a facility —
rendered across 45+ pages.

The visual system carries one positioning idea across every surface: warm photography over claims, real names and faces, dense trust signals, identical templates so 45+ pages feel like one product.

Homepage. Trust signals first: govt-registration line, real photography, two-tap CTAs (Schedule a Visit / Call). Photography brief was given to the founder so every image was a real moment, not stock.

Services overview. 12 specialized care types — Senior Citizen, Alzheimer’s, Paralytic & Stroke, Cancer Care, Post-Surgery Recovery, NRI Elder Care plus six more. Each card maps to a pillar SEO page.

Service detail (pillar). Identical template enforced by Claude Code skill 03 — same heading hierarchy, same checklist density, same CTA placement. Repeatable across all 12 services.

Daily routine. The most-asked question from adult children is ‘what does a day look like?’ The schedule answers it before they ask. A footnote acknowledges medical adjustments — honesty over polish.

Our Story. Sammeer and Ssunita Nawathe on the page with names, faces, and the reason. Trust on an emotional decision is built by people, not logos.

Facilities. The room a parent will sleep in — shown, not described. Adult children compare physical environment first; copy second.

Articles index. The visible surface of the pillar-cluster system — guides indexed by category.

Article detail. English / Hindi / Marathi toggle on every guide. Key Takeaways block, on-this-page nav, and dense internal linking back to the pillar service page.

Design decisions

Eight decisions that did
the disproportionate work.

Every decision maps to a strategic insight, not an aesthetic preference. The Claude Code skills are the operating manual; these are the principles that shaped them.

Context

Ranking, intent-rich, engineered clean.
The system is doing the work it was designed to do.

Search Console · 3 months

28 clicks, 1.41K impressions, avg position 5.8 from a brand-new domain. Compounding curve, exactly the shape pillar-cluster produces.

Top queries

124 unique ranking queries. Brand terms plus generic intent terms (‘old age home in borivali west’, ‘best old age home in mumbai’) — the cluster strategy capturing both.

PageSpeed Insights

97 / 94 / 100 / 100 with LCP 1.1s on a mobile-first audience. Engineering quality matches the design quality. Bootstrap + static HTML wasn’t the lazy choice — it was the right one.

Closing

One person, five days, a SEO-ranked product.
The agency quote was ₹1.5 lakh and 6 weeks.

That gap is what designing in code, with judgment encoded as systems, looks like in practice.

Deep dive

For the reader who wants the full story.

Chapter 1 · Discovery

How 4 hours produced the positioning strategy.

Four hours of structured conversation with the founder, Claude and ChatGPT used as synthesis tools. The goal: understand the real problem before touching a design tool. The primary user is an adult child, 40–65, making one of the most emotionally charged decisions of their life. They are not browsing — they are in a moment of acute concern.

Chapter 2 · Competitive analysis

Borivali elder care — quality bar wide open.

The 1–2 stronger players had invested in content and SEO but still felt institutional — clinical, dated, cold. Aannapurnaa Aai couldn’t outspend a multi-building campus. It could out-feel them. A modern site that felt like home would stand out in a market where everything else felt like a hospital admission form.

Chapter 3 · Claude Code skills

Project-scoped guardrails for an AI engineer.

Claude Code skills are markdown files (SKILL.md) that live inside the project repo. Claude Code automatically loads any matching skill before it executes a task — design changes, content writing, page generation. Each skill is a contract: when Claude touches this file pattern, it must obey these rules.

Each one I wrote followed the same design pattern: identify a failure pattern Claude would hit at scale → encode the rule with concrete examples → ship it before any production code touched the repo.

What changed: 45+ pages had identical heading hierarchy, identical CTA placement, identical tone, identical SEO heading patterns. Not because I reviewed each one — because Claude Code couldn’t violate the skills even if it tried. Consistency was a guarantee, not a hope.

Chapter 4 · DataForSEO

Research-first content, not retro-fitted SEO.

For every blog topic, Claude queried DataForSEO for ranking keywords (short-tail and long-tail), search volume, competitor content, and trending queries in the elder care space. Articles were designed from the keyword data up — not blogs that were tagged after writing. Every piece of content was both expert-quality and rank-optimized by design.

Chapter 5 · Pillar + cluster architecture

Not a blog — an SEO architecture.

Each of the 12 services became a pillar topic. 2–3 cluster articles were built around each pillar, every cluster linking back to its pillar and across to related clusters. Total: 30–35 articles operating as an interconnected ranking system.

Chapter 6 · The Hour-2 quality gate

Test the skills on 2 pages before letting them
ship 45.

The quality gate is the moment between writing the Claude Code skills and trusting them. At Hour 1, I had Claude generate 2 throwaway pages using the skills as written. At Hour 2, I reviewed those pages line-by-line. Three failure patterns appeared — and each one was a gap in the skills, not in Claude.

Each fix went back into the relevant Claude Code skill before any production page was generated. Consistency rate after the fixes: ~95% across all 45+ subsequent pages, no per-page review cycle needed.

Chapter 7 · Tech stack

The right tool for the actual problem —
not the most impressive one.

Three real options for a 45-page informational site, mobile-first audience, non-technical owner, ₹0 hosting budget. Scored against what the project actually needed.

The site is static and informational. Speed and lightweight delivery matter for the mobile users searching on the go. No CMS was needed since updates flow through me. Adding Next.js would have been engineering for engineering’s sake; choosing WordPress would have inherited 500KB of plugin bloat and a SQL database for a site that needed neither. Framework selection is a product decision, not a technical one.

The receipt: 97 / 94 / 100 / 100 with LCP 1.1s on desktop, sub-2s on 4G mobile.

Chapter 8 · What I’d do differently

Always instrument before you ship.

I’d also consider a lightweight CMS — even a Notion-to-HTML pipeline — so the founder could make basic updates without me. The current setup works because I’m available, but that’s a dependency, not a solution.

Chapter 9 · What this taught me about AI-assisted development

Claude Code skills aren’t efficiency tools.
They’re quality control systems.

The mistake most professionals make with AI-assisted development is treating Claude Code as a faster typist. It is not. It’s a junior engineer that will execute literally whatever it’s asked — and at scale, that produces inconsistent output. Claude Code skills (project-scoped SKILL.md files) are how you stop being a prompter and start being an architect.

The 3-step methodology that came out of this project: (1) write the Claude Code skills first → (2) run an Hour-2 quality gate to find the gaps → (3) patch the skills, then let the agent ship at scale. Three days of thinking before six hours of building is how you produce quality at speed.

Pro bono work, executed at the same standard as commercial work, produces portfolio pieces just as strong. The client’s budget does not determine the quality of the thinking.