Own Product
felixschmidt.software — AI-Powered Portfolio
Personal portfolio and marketing website with an integrated AI assistant powered by Retrieval-Augmented Generation (RAG). Visitors can ask natural-language questions about Felix's experience, and the system retrieves relevant projects via vector similarity search before generating grounded, hallucination-free answers using Google Gemini. The entire pipeline is bilingual (EN/DE), streaming, and built on a modern full-stack architecture.
Die Herausforderung
Portfolio websites suffer from a discovery problem: visitors have specific questions about a freelancer's experience, but the answers are scattered across project cards, tech lists, and case studies. Traditional portfolios force visitors to navigate, scan, and piece together information — leading to high bounce rates and missed opportunities. Additionally, generic AI chatbots hallucinate project names, fabricate technologies, and describe experiences that never happened, making them unsuitable for professional credibility.
Der Ansatz
Designed and built a purpose-built RAG (Retrieval-Augmented Generation) system from scratch. Project data is stored in Supabase PostgreSQL with bilingual descriptions (EN/DE). Each project is converted into a 768-dimensional vector embedding using Google Gemini Embedding, stored via the pgvector extension. When a visitor asks a question, the query is embedded with the same model and matched against project embeddings using cosine similarity search. The top matching projects are injected as context into a Gemini 2.5 Flash prompt, which generates a streaming, grounded response. The system includes lead routing with buy-signal detection, a Project Fit Analyzer for detailed project matching, and a full admin panel for embedding synchronization. The frontend uses React 19 with CSS Modules, Framer Motion animations, and Three.js for 3D elements. Infrastructure runs on Vercel (frontend) with Railway (NestJS backend) and Supabase (database).
Tech Stack
Ergebnisse & Wirkung
The AI assistant enables visitors to find relevant information about Felix's experience in seconds instead of browsing through project pages. The RAG system ensures every answer is grounded in real project data with zero hallucinations. Streaming responses provide immediate feedback, and bilingual support serves both English and German-speaking visitors natively. The Project Fit Analyzer provides potential clients with a structured assessment of how Felix's skills match their specific requirements, streamlining the pre-sales qualification process.
Meine Rolle
Solo developer and architect. Responsible for the full product — from concept and architecture design through implementation of the RAG pipeline, NestJS backend, React frontend, Supabase schema, embedding generation, vector search, AI prompt engineering, infrastructure setup (Vercel, Railway), and deployment automation.
Ähnliche Herausforderung?
Erzähl mir von deiner Herausforderung. Ich sage dir, wie ich sie angehen würde — und eine realistische Timeline.