this Portfolio AI

Portfolio with text to portfolio that can text

My portfolio used to be just still pages. I wanted visitors to chat with it and dig into any project on their own.

The Challenge

• Teach myself RAG, vector databases, and orchestration from scratch
• First offline build with Ollama + Llama maxed out my laptop after a single chat
• Find hosting that could juggle 30+ chats (atleast), keep history, and stay basically free
• Wrap it all in a simple yet scalable and elegant UI that fits into my portfolio

That's the hill I had to climb; the next section will show how I tackled it.

My Role: Solo Design Engineer & AI Architect

As the sole creator, I owned the entire lifecycle, demonstrating my ability to manage strategy, architecture, and execution.

• Strategy: I defined the project goals and made the critical pivot from a self-hosted to an API-driven architecture
• AI Architecture: I designed the end-to-end RAG system using Gemini APIs, Qdrant, and the Haystack framework
• Full-Stack Development: I built the Python backend, integrated all services, and connected it to my Next.js portfolio frontend
• Infrastructure & Deployment: I deployed and optimized the application on Oracle Cloud, engineering it to work within the constraints of the free tier

Process: From Idea to Live AI Chat

Back in March 2025 I wondered, “What if my portfolio could actually talk back?” With zero budget and just my laptop, I decided to find out.

First local test — I used Ollama, Llama 3.2, Chroma, and LangChain. It worked—until one answer swallowed 6 GB of RAM and froze my MacBook. Fun demo, unusable in real life.

Learning the basics — Evenings went into reading about retrieval-augmented generation (RAG), vector databases, and prompt design. ChatGPT and Claude filled gaps while I sketched ideas on paper.

Moving to APIs — Heavy models had to go. I picked Qdrant’s free tier for vectors, Gemini Flash for replies, and OpenAI embeddings (5 million tokens for $5). Swapped LangChain for Haystack because the code felt cleaner.

Free-tier hosting — I set up an Oracle Cloud VM, served the backend with uvicorn, added health checks and session cleanup so the free tier wouldn’t shut me down. Target: 30+ chats at once without breaking.

Hooking up the front end — Vercel v0 and Next.js gave me a quick frontend shell of portfolio. Mm FE skills helpt me engineer it. Dropped in the chat widget, tweaked the styles, and connected it to the API.

Polish and test — Logs, retries, and late-night bug fixes followed. Now the chatbot runs lean, stays free, and lives right here in my portfolio. But, next steps are mapped out and are as such: refinig the answers, enhancing content pipeline and implementing your feedback.

Outcomes & Impact: A Smarter, More Engaging Portfolio

This project successfully transformed my portfolio into an interactive tool and a powerful demonstration of my capabilities (did it?).

Enhanced User Engagement: Visitors would no longer be passive viewers. They can now actively inquire and receive tailored information, turning a monologue into a dialogue (do they?).

Demonstrated Strategic Adaptability: The documented pivot serves as clear evidence of my ability to diagnose a failing strategy and make pragmatic trade-offs to deliver a superior, more sustainable solution.

Proven Technical Acumen: The final application successfully handles up to 30 concurrent user requests with efficient response times, proving my ability to build and deploy scalable AI systems.

Future-Ready Skills: This project is a tangible showcase of my proficiency in the core components of modern AI applications (LLMs, Vector Databases, RAG), positioning me at the intersection of design and AI engineering.

Technologies & Frameworks

• AI/ML: RAG, Gemini LLM, OpenAI embeddings, Splade Sparse Embeddings, Ollama + Llama
• Orchestration: Haystack, Langchain
• Database: Qdrant Vector Database, Chroma
• Infrastructure: Oracle Cloud
• Frontend: Next.js, Vercel v0, React, TypeScript
• My AI Teachers/Tools: ChatGPT, Perplexity, Claude, Co-pilot

This project demonstrates my commitment to staying current with AI technologies while combining design thinking with engineering implementation—essential skills for a Design Engineer role in today's AI-driven product landscape.