The problem
Most professionals have a LinkedIn profile but no owned space that reflects how they actually think. A static CV lists titles and dates. It doesn't show frameworks, perspective, intellectual range, or what someone is like to work with.
The alternative — hiring a developer, buying a template, learning a web framework — felt like solving the wrong problem. The real challenge was not technical. It was deciding what to say and being specific enough that it sounds like you, not a polished version of you.
The site should give before it asks. Every visitor should walk away with something useful, even if they never reach out.
The guiding principle
The site had to do three things well:
- Reflect genuine thinking — not a LinkedIn summary, but actual frameworks, real opinions, and honest perspective on AI and product management.
- Give value on arrival — a calendar with 30 AI concept cards and a 12-step Beginner's Journey means every visitor learns something, whether or not they ever contact you.
- Stay maintainable — adding a new project, article, or travel destination should take one conversation, not an afternoon of editing code.
How it was built — milestones
Every milestone was a conversation. The role division was consistent throughout: direction, content, and decisions came from one side — execution came from the other.
Milestone 01
Site live in 60 seconds
Described goals, field, and creative vibe. Claude built a complete HTML file. Uploaded to GitHub Pages — live at mdevi25.github.io, zero cost, zero domain purchase.
You: direction
Milestone 02
Real content integrated
Travel photos uploaded by filename. Two Medium articles shared by URL — titles, subtitles, and read times fetched and built into cards. LinkedIn, GitHub, Medium contact links added.
You: content
Milestone 03
Two project case studies
Uploaded raw .md documents. Claude produced full standalone pages — each with a table of contents, pillar cards, metric blocks, decision diagrams, and a back-to-site nav.
You: documents
Milestone 04
Perspective widget
Defined worldview, PM philosophy, and AI stance using a structured template. Claude built a widget where visitors ask questions and receive answers grounded in your actual frameworks — not a generic AI response.
You: thinking
Milestone 05
Calendar with AI insights
Approved concept for seasonal animated birds and AI concept cards. Claude integrated the full calendar — 30 AI concept cards, a 12-step Beginner's Journey, and an expandable read panel.
You: concept approved
Milestone 06
Iterative polish
Contact layout, visit counter, expand-to-read panel, text edits, colour tweaks — every refinement was a single sentence instruction. No code touched manually at any point.
You: decisions
Code written manually
0 lines
The stack
The entire build runs on free infrastructure. No APIs, no backend, no recurring costs. The site is entirely static HTML served by GitHub Pages.
You — product brain
Layer 1 — thinking
Madhu Devi
Goals, content, decisions, approvals. Every direction came from here. Claude executed — you decided.
Product brain
Claude AI — execution layer
Layer 2 — AI partner
Claude (Anthropic)
HTML, CSS, JavaScript, page design, content formatting, and integration logic — all generated through conversation. Also powers the live Perspective widget on the homepage.
claude-sonnet-4
Layer 3 — interface
Claude.ai
The conversation interface. No IDE, no terminal, no code editor used at any point.
chat interface
Infrastructure — free tier
Layer 4 — hosting
GitHub Pages
Free static hosting. Live at mdevi25.github.io with automatic deployment on file upload.
$0 / month
Layer 5 — source
GitHub repo
4 HTML files + 1 images folder. No build process, no dependencies, no framework.
4 files
Layer 6 — domain
GitHub subdomain
mdevi25.github.io — free, no domain purchase, no DNS configuration needed.
no cost
$0 recurring
No subscriptions. No hosting fees. No domain purchase. Built and maintained entirely through conversation.
Key product decisions
Decision 1
GitHub Pages over Webflow, Framer, or Squarespace
No domain purchase, no subscription, no vendor lock-in. GitHub Pages gives a permanent free URL and keeps all files in your control. The constraint of a single HTML file turned out to be a feature — simplicity forces clarity.
Decision 2
Conversation over prompts
Every change was an iterative conversation, not a one-shot prompt. This is the key insight: AI collaboration works best when you treat it like working with a capable colleague — context, feedback, and refinement over time.
Decision 3
A hidden gem over a contact form
The calendar with 30 AI concept cards and a 12-step Beginner's Journey was a deliberate product decision — give visitors something useful on arrival. A site that teaches earns more trust than a site that only asks.
Decision 4
Perspective widget over a generic chatbot
Rather than a generic AI assistant, the widget is grounded in actual frameworks, real writing, and defined opinions. Visitors get a genuine point of view — not a search engine or a placeholder.
Key takeaways
1
The hardest part is content, not code. Deciding what to say — and being specific enough that it sounds like you — is where the real work is. AI handles the execution. The thinking is yours.
2
AI collaboration is a product skill. Knowing what to ask for, when to push back, when to approve, and when to redirect — that is product management applied to your own build. The output quality mirrors the quality of your direction.
3
Specificity is everything. Vague inputs produce generic outputs. The moments where the site feels most like me are the moments where I gave Claude the most specific brief — actual frameworks, real opinions, exact wording preferences.
4
A living portfolio beats a static CV. The site continues to grow — new projects, new articles, new travel destinations — each added in a single conversation. This is the right model for a professional in a fast-moving field.