← All case studies

Case Study · Santosh Paudel (Personal)

Building a Full-Stack AI Portfolio OS in 25 Days

25

Days to Ship

24

Blog Posts Seeded

4

Platforms Shipped

70

SEO Pages Generated

!The Challenge

Most portfolio sites show what someone claims to do. This one had to show it — with live AI infrastructure, autonomous agents, a client CRM, and programmatic SEO — all built solo in a 25-day sprint.

The Strategy

Next.js 15 + Supabase + Vercel full-stack build. Autonomous Claude API agents with tool-use loops, approval workflow, and persistent memory. Admin CRM + HRM + POS panels. 60+ programmatic SEO routes for industries and local pages. 24-post blog seeded via SQL. All deployed in 25 days using Claude Code.

The Results

Four production platforms live. AI agents running the daily brief autonomously. 24 blog posts indexed. 70+ programmatic SEO pages generating organic traffic. Admin panel managing clients, staff, and sales from one codebase.

The Challenge

A portfolio site is supposed to prove you can do the work. For a developer building AI systems and full-stack platforms, that means showing — not telling. The challenge: build a production portfolio that itself demonstrates autonomous AI agents, programmatic SEO, full-stack admin panels, and modern dev automation.

Time constraint: 25 days. No team. Just Claude Code, Next.js, and a clear vision.

The Stack

  • Next.js 15 with React 19, TypeScript, App Router
  • Supabase for PostgreSQL, auth, RLS policies, and real-time
  • Vercel for deployment, CI/CD, and ISR edge caching
  • Claude API for the autonomous agent system
  • Groq (Llama 3.3 70B) for fast agent inference
  • DeepSeek R1 for complex reasoning tasks

The Build

Phase 1: Core Platform (Days 1–8)

Built the public portfolio with blog, services, case studies, and contact. Implemented ISR (revalidate: 300). Set up Supabase schema with RLS policies. Deployed to Vercel on day 3.

Phase 2: Admin CRM (Days 8–14)

Full admin panel with Supabase auth, client pipeline (lead → prospect → proposal → active), task management, content planner, and email subscriber management. Mobile-first with bottom tab navigation.

Phase 3: Agent System (Days 14–20)

Autonomous Claude API agents with tool-use loops. Agents read business context, propose blog drafts and client tasks, log to pending_actions, and wait for human approval. Never write directly to production tables. Daily brief runs via Vercel Cron at 9am weekdays.

Phase 4: Programmatic SEO + Content (Days 20–25)

60+ industry pages and 20+ local pages using generateStaticParams. 24 blog posts seeded via SQL with dollar-quoted markdown. Schema markup on every page. llms.txt for AI search crawlability.

Results

Four production platforms shipped: the portfolio OS, the news platform (TheWestNepal.live), the AI agent marketplace (agenticai01.tech), and a US gaming client backend — all within the 25-day sprint.

The AI agent system runs the daily brief autonomously, proposes content, and surfaces client action items — all reviewable from the admin panel. Zero rogue actions since launch.

Key Learnings

Claude Code compressed development time by ~60% versus traditional coding. The approval workflow pattern (agents propose, humans approve) is the right mental model for autonomous AI in production. Supabase + Next.js + Vercel is the modern founder stack — zero infra ops, scales to millions.

ShareLinkedInX / Twitter

External Resources

Research & Benchmarks Behind These Results

Ready for similar results?

Let's talk about your situation — no pitch, just a conversation.

Loading calendar…

or view all services →