Creative Technologist & AI Agent Developer
I design and build production-grade AI agent systems — from single-agent RAG pipelines to multi-agent orchestration frameworks. My work spans agentic workflow design, retrieval-augmented generation, prompt engineering, full-stack AI applications, and enterprise AI architecture. I also apply generative AI tools and prompt engineering techniques to produce commercial brand imagery for major retail clients.
Based in US & EU/Ireland.
🌐 paulforlando.com | 💼 LinkedIn | 📧 Available for freelance & consulting
Single Agents → Multi-Agent Systems → Enterprise Orchestration Pipelines → Full-Stack AI Applications
I focus on agents that are production-ready — properly configured, defensively prompted, and designed to fail gracefully. Not just demos.
| Agent | Pattern | Stack | Demo |
|---|---|---|---|
| Food Chatbot App | Agentic RAG + Cart | Next.js · FastAPI · ChromaDB · OpenAI | 🔗 Live |
| AI Agent Team Supervisor App | Supervisor Pattern | OpenAI Agents SDK · Next.js · FastAPI · ChromaDB | 🔗 Live |
| Data Analysis Agent App | Interactive Data Agent | Claude Code · Next.js · FastAPI · OpenAI · Recharts | 🔗 Live |
| Deep Research Agent App | Full-Stack Research App | Claude Code · Next.js · OpenRouter · Exa AI · TypeScript | 🔗 Live |
| Web Research Hub | Hierarchical 3-Agent Pipeline + MCP | Next.js · FastAPI · OpenRouter · Gemini 2.5 Flash · Exa AI · MCP | 🔗 Live |
| Web Research Hub MCP Server | Custom MCP Server · Research Tools | FastAPI · FastMCP · Streamable HTTP · Exa AI · Python | 🔗 Live |
| GenAI Concepts Chat | Agentic RAG + Custom MCP Server | Node.js · Express · TypeScript · Pinecone · OpenRouter · Gemini Flash 2.5 | 🔗 Live |
| Pinecone Agentic Search MCP Server | Custom MCP Server · Agentic RAG | Node.js · TypeScript · Pinecone · OpenRouter · MCP Protocol · Railway | 🔗 Live |
| AI Document Generator | LLM Chain + Quality Gate | n8n · OpenRouter · GPT-4.1 · LangChain | — |
| AI Agent Team — Supervisor Pattern | Supervisor Orchestration | Flowise AgentFlows V2/V3 · GPT-4o · LangChain | — |
| AI Food Chatbot Agent | Agentic RAG + Tool Routing | Flowise · GPT-4o · Postgres · OpenAI Moderation | — |
| AI Multi-Agent Content Pipeline | Sequential Multi-Agent | Flowise · GPT-4o · FAISS · RAG | — |
| AI Web Research Agent | RAG + Web Scraping | Flowise · GPT-4o-mini · FAISS · Cheerio | — |
| AI Research Assistant RAG | Lightweight RAG | Python · OpenAI · NumPy · Scikit-learn | — |
| Data Analysis Agent | Custom GPT | GPT-4 · Python · Pandas · Scikit-learn | — |
I use generative AI tools with prompt engineering techniques to produce brand imagery for major retail clients across the following disciplines:
- Generative AI Image Creation
- AI Art Direction
- Commercial Product Visualization
- Lifestyle Imagery
🔗 View Portfolio on ArtStation | Repository
Agent Design — tool routing, prompt engineering, multi-agent orchestration, supervisor patterns, retrieval-augmented generation, hallucination detection, moderation, memory, full-stack AI applications, MCP server development
Stack — Flowise · LangChain · OpenAI API · Python · FastAPI · Next.js · n8n · TypeScript · OpenRouter · Exa · Postgres · FAISS · Neon · Supabase · Claude Code · Pinecone · FastMCP · MCP Protocol · Railway · Vercel
Disciplines — 3D Visualization · Generative AI · Data Analytics · AI Product Visualization
Every agent in this portfolio is built with the same standard:
- Explicit, rule-based system prompts — no vague instructions
- Tool descriptions written as policies, not labels
- Temperature tuned to the use case — not left at default
- Failure modes addressed — iteration caps, moderation, fallbacks
- Production considerations documented — memory, security, deployment
Every live application in this portfolio is built with production-grade security and cost controls — not just functional demos.
MCP Server Security
Both custom MCP servers implement API key authentication (X-API-Key header, 401 on invalid key) and sliding-window rate limiting (5–10 requests/IP/hour, 429 on exceed) with self-host instructions embedded in every error response. Rate limiting is implemented as pure middleware without third-party auth frameworks — correct IP detection behind Railway's proxy via X-Forwarded-For header parsing.
Cost Protection All LLM API keys (OpenAI, OpenRouter) are capped at hard monthly spend limits. Exa AI auto-recharge is capped per calendar month. Rate limiting at the infrastructure layer provides the first line of defense; spend caps at the provider level provide a hard ceiling if rate limiting is ever bypassed.
Production AI systems require controls at every layer — request-level rate limiting, infrastructure-level authentication, and provider-level spend caps. Each application in this portfolio is built with all three, reflecting the same standards applied in enterprise deployments where cost, security, and reliability are non-negotiable.
Open to collaboration on agent design, AI workflow architecture, and creative technology projects.