AI engineer with 15+ years in software, writing about taking AI from prototype to production.https://www.raaulc.com/https://www.raaulc.com/posts/spotify-agent-router/https://www.raaulc.com/posts/spotify-agent-router/Spotify put a bouncer in front of its coding agent — a rule that stops the expensive model from reading big files and hands the job to a cheap one. Here is how the router works, and what they deliberately left alone.Sun, 06 Sep 2026 00:00:00 GMThttps://www.raaulc.com/posts/chat-with-docs-langchain-rag/https://www.raaulc.com/posts/chat-with-docs-langchain-rag/Build a document Q&A system using LangChain, OpenAI, and FAISS. Load PDFs, chunk them, embed into vectors, and ask questions with conversational memory — the full RAG pipeline in ~30 lines.Wed, 01 Apr 2026 00:00:00 GMThttps://www.raaulc.com/posts/context-hub-vs-context7/https://www.raaulc.com/posts/context-hub-vs-context7/How Context Hub and Context7 fix the silent bug that every AI coding assistant has.Mon, 09 Mar 2026 00:00:00 GMThttps://www.raaulc.com/posts/self-rag-pipeline/https://www.raaulc.com/posts/self-rag-pipeline/Build a self-evaluating RAG pipeline with LangGraph — the LLM grades its own retrieved documents, checks for hallucinations, and validates its answers before returning them.Wed, 11 Feb 2026 00:00:00 GMThttps://www.raaulc.com/posts/quiz-generator-with-llama/https://www.raaulc.com/posts/quiz-generator-with-llama/Build an MCQ quiz generator in ~100 lines — paste any text, pick a difficulty, and LLaMA 4 via Groq returns 5 multiple choice questions as structured JSON. Auto-scoring included.Sun, 18 Jan 2026 00:00:00 GMThttps://www.raaulc.com/posts/ai-anime-chatbot-with-rag/https://www.raaulc.com/posts/ai-anime-chatbot-with-rag/Build an anime chatbot that scrapes any URL, embeds it with FAISS, augments with DuckDuckGo search, and answers via Groq's Llama 3.3. Full RAG pipeline with source citations.Mon, 22 Dec 2025 00:00:00 GMThttps://www.raaulc.com/posts/variational-autoencoders-mnist/https://www.raaulc.com/posts/variational-autoencoders-mnist/Build a Conditional VAE in PyTorch that generates handwritten digits on demand. Covers the reparameterisation trick, KL divergence loss, and conditional generation.Tue, 14 Oct 2025 00:00:00 GMThttps://www.raaulc.com/posts/fine-tuning-gpt2-peft-lora/https://www.raaulc.com/posts/fine-tuning-gpt2-peft-lora/Fine-tune GPT-2 on English quotes using LoRA adapters — only 0.24% of parameters trained. Covers LoRA theory, PEFT setup, FP16 training, and inference.Tue, 08 Jul 2025 00:00:00 GMThttps://www.raaulc.com/posts/finetune-t5-small-summarization/https://www.raaulc.com/posts/finetune-t5-small-summarization/Fine-tune T5-Small on the SAMSum dialogue dataset using HuggingFace Transformers. Covers tokenization, Trainer API, gradient accumulation, and inference with pipeline.Sun, 20 Apr 2025 00:00:00 GMThttps://www.raaulc.com/posts/next-word-prediction-lstm/https://www.raaulc.com/posts/next-word-prediction-lstm/Build a next word prediction model from scratch using an LSTM in PyTorch, trained on Medium article titles. The same core idea behind every language model.Wed, 15 Jan 2025 00:00:00 GMT