"UGLE" AI Office Assistant — &SHUFL Kitchen Design (Prototype)

Designed and built a prototype AI-powered office assistant for a 12-person kitchen design firm, integrated directly into their existing internal communication workflow via Slack. The assistant gives staff instant access to product knowledge, pricing, meeting notes, and reference materials through natural language — eliminating manual searches across disconnected internal systems.

What was built:
A RAG-powered chatbot deployed on Google Cloud, combining a vector knowledge base of company-specific product data (the &SHUFL collection) with an LLM for natural language response generation. The assistant retrieves relevant documents from internal databases — including technical specs, price lists, meeting summaries, and reference images/drawings — and synthesises accurate, context-aware answers in conversational Danish or English.

Key responsibilities and achievements:Designed the full RAG architecture: document ingestion, chunking strategy (384-token chunks, 50-token overlap), vector embedding, similarity retrieval, and answer generation
Integrated Slack API as the user-facing interface, meeting the team where they already work
Connected OpenAI API for LLM-powered generation and ChromaDB as the vector store
Conducted systematic benchmarking against competing solutions (Google NotebookLM, OpenAI Custom GPT) to diagnose retrieval failures and optimise parameters
Tuned retrieval thresholds and context window management to improve answer precision
Deployed on Google Cloud Platform

Technologies and platforms:

LLM: OpenAI API (GPT)
Vector database: ChromaDB
Interface: Slack API
Infrastructure: Google Cloud Platform
Architecture pattern: Retrieval-Augmented Generation (RAG)
Languages: Python