Steve Arnold Otieno — Portfolio v3.0 (Digital Twin AI)

v3.0 represents a fundamental shift — from a static portfolio to an Autonomous Technical Twin. Built on an enterprise-grade speed stack, it features a proactive AI agent that conducts real-time technical screenings, explains architectural decisions, and surfaces direct evidence from my GitHub, LinkedIn, and live production environments.

➡️ View Live Agent & Portfolio

---

✨ Engineering Highlights (V3 Evolution)

• Proactive AI Technical Twin: A bi-directional agent powered by Groq (Llama 3.3) and WebSockets. It monitors user intent and proactively offers contextual deep-dives — e.g. "I see you're looking at GradeHub; want to know how I handled multi-tenant RBAC?"
• Enterprise RAG Pipeline: A custom Retrieval-Augmented Generation engine backed by Neon (PostgreSQL + pgvector), performing semantic search across my entire professional history for 100% factual accuracy.
• Automated Knowledge Ingestion: A modular ETL pipeline that auto-syncs from:
  • GitHub — real-time README scraping and technical metadata extraction
  • LinkedIn — standardised career history via PDF-parsing
  • Live Web — recursive crawling of production sites for brand consistency
• Local Inference Engine: High-performance vector embeddings generated on-server with Xenova Transformers (all-MiniLM-L6-v2) — zero API cost, zero-latency vectorisation.
• The Shield (Rate Limiting): Context-aware throttling via NestJS Throttler protecting both HTTP and WebSocket layers from bot spam.

---

🛠️ Tech Stack

Frontend

Layer	Technology
Framework	Next.js 15+ (App Router)
Real-time UI	Socket.io-client & Framer Motion
Rich Media	React Markdown with GFM support
Design	Tailwind CSS & shadcn/ui
Forms	React Hook Form & Zod
Email	Resend & React Email
Icons	Lucide React
Notifications	Sonner

Backend

Layer	Technology
Framework	NestJS (Modular Architecture)
AI Inference	Groq Cloud LPU — Llama 3.3 70B
Vector DB	Neon with pgvector
ORM	Drizzle ORM (type-safe migrations)
Ingestion	Playwright (headless scraping) & pdf-parse

Infrastructure

Layer	Technology
Containerisation	Docker (multi-stage builds)
Hosting	AWS EC2 (t3.micro) + Nginx reverse proxy
CI/CD	GitHub Actions — automated deployment & model pre-caching
Frontend Deploy	Vercel

---

📂 Repository Structure

steve-portfolio-v3/
├── client/                 # Next.js Frontend → Vercel
│   ├── src/hooks/          # Custom AI & Socket hooks
│   └── src/services/       # Singleton Socket management
├── server/                 # NestJS Backend → AWS EC2
│   ├── src/modules/        # Modular AI, Vector, and Chat domains
│   ├── src/scripts/        # Knowledge seeding and ETL scripts
│   └── Dockerfile          # High-performance production image
└── infra/                  # Nginx & SSL configurations

---

🚀 Running Locally

This is a monorepo managed with pnpm.

1. Clone the repository:

git clone https://github.com/Arnoldsteve/Portfolio.git
cd steve-portfolio-v3

2. Server setup:

cd server
pnpm install
# Add GROQ_API_KEY and DATABASE_URL to .env
pnpm db:push     # Run Drizzle migrations
pnpm seed        # Initialise AI memory
pnpm start:dev

3. Client setup:

cd client
pnpm install
# Add RESEND_API_KEY to .env.local
pnpm dev

Open http://localhost:3000 to see the result.

---

☁️ Deployment

• Frontend: The client/ directory deploys automatically to Vercel on every push to main.
• Backend: The server/ directory is containerised via Docker and deployed to AWS EC2 through a GitHub Actions CI/CD pipeline.

---

📜 License

This project is licensed under the MIT License. See the LICENSE file for details.

This project follows Conventional Commits and SOLID design principles. For an interactive technical deep-dive, talk to the AI Twin on the live site.

---

📬 Contact

Steve Arnold Otieno — Solutions Architect & Full-Stack Engineer

• Email: stevearnold9e@gmail.com
• LinkedIn: linkedin.com/in/steve-arnold-otieno