I built a RAG chatbot in 45 minutes. No coding. It's a great way to learn or create a solution for an AI PM portfolio. But how does it work? First, RAG isn't just about vector stores. It might involve retrieval from any data source like Google Drive, SQL, or text files. I used a vector store, so I needed to: 𝐒𝐭𝐞𝐩 𝟏: 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐞 𝐄𝐦𝐛𝐞𝐝𝐝𝐢𝐧𝐠𝐬 The data isn’t stored in its original format. Instead, it's split into chunks (e.g., 500-1000 characters each), which are then converted into multi-dimensional vectors and stored in a vector database. 𝐒𝐭𝐞𝐩 𝟐: 𝐇𝐚𝐧𝐝𝐥𝐞 𝐑𝐞𝐭𝐫𝐢𝐞𝐯𝐚𝐥, 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐨𝐧, 𝐚𝐧𝐝 𝐔𝐈 The users interact with a Lovable interface. When someone asks a question, it's turned into a vector and used to retrieve the most relevant chunks. Finally, an LLM uses retrieved chunks and the original request to generate an answer. That's Vanilla RAG - the simplest possible implementation. In practice, consider: • 𝐀𝐝𝐚𝐩𝐭𝐢𝐯𝐞 𝐑𝐀𝐆: Dynamically selects or modifies the retrieval strategy (e.g., choosing the best data source, reformulating the query) based on the request. • 𝐇𝐲𝐛𝐫𝐢𝐝 𝐑𝐀𝐆: Combines multiple retrieval approaches, e.g., keyword + semantic search, or merges results from multiple sources. 𝐒𝐭𝐞𝐩 𝟑: 𝐄𝐯𝐚𝐥𝐮𝐚𝐭𝐞 𝐑𝐀𝐆 RAG systems have two distinct components that require different evaluation approaches: retrieval and generation. Traditionally, we evaluate the retrieval part using metrics like Recall@k, Precision@k, or MRR. I loved Jason Liu’s post "There Are Only 6 RAG Evals." He noticed that RAG systems have 3 core components: • A question (Q) • Retrieved context (C) • An answer (A) In his post, he explains how to evaluate each of those combinations: https://lnkd.in/dGgcqKB5 𝐓𝐡𝐞 𝐑𝐞𝐜𝐨𝐦𝐦𝐞𝐧𝐝𝐞𝐝 𝐓𝐞𝐜𝐡 𝐒𝐭𝐚𝐜𝐤 You can build it virtually for free: • UI: Lovable (free version is enough) • Orchestration: n8n (trial / free self-hosted edition) • LLM: GPT-4o-mini by OpenAI (less than $2 for 100s of requests) • Embedding model: text-embedding-3-small • Vector database: Pinecone (free tier, Starter) • Documents' source: Google Drive --- In my post, I explain how to build a RAG chatbot step-by-step. No coding. This might help you: • Learn by doing • Develop a better AI intuition • Build an AI chatbot for your portfolio Demo, templates, and more with a free trial: https://lnkd.in/dew--RqD --- P.S. Working on AI products? I highly recommend AI Evals for Engineers and PMs cohort by Hamel Hussain and Shreya Shankar. It's the No. 1 course on Maven right now. I've just completed it together with 700+ other students (e.g., Teresa Torres who left a testimonial). It helped me understand why so many AI products fail and what successful teams do differently. The 𝐥𝐚𝐬𝐭 𝐥𝐢𝐯𝐞 𝐜𝐨𝐡𝐨𝐫𝐭 starts on July 21. A special $𝟖𝟎𝟎 𝐝𝐢𝐬𝐜𝐨𝐮𝐧𝐭 code for our community: https://bit.ly/aievals35 | 20 comments on LinkedIn