# largo.dev > Independent applied AI engineering notebook by Steven W. White featuring first-principles architecture deep dives, measured GPU and agent benchmarks, and complete runnable code. Every guide on largo.dev pairs architectural intuition with reproducible benchmarks across multi-agent orchestration (Google ADK, LangGraph, AWS Strands), stateful agent memory (SQLite vector RAG, Vertex AI Memory Bank), open-weight MoE serving (Gemma 4 LoRA, vLLM on Vertex AI), two-stage hybrid retrieval (BM25, dense embeddings, RRF, and cross-encoder reranking), PyTorch encoders, DeepSeek reasoning internals, and production ML serving. ## Core Index, Tools & Feeds - [Full Technical Corpus for LLMs (llms-full.txt)](https://largo.dev/llms-full.txt): Complete Markdown text, code blocks, and benchmark tables for all 27 tutorials and 5 essays. - [Interactive AI Systems Engineering Lab](https://largo.dev/lab/): Three browser-based simulators covering (1) GPU VRAM, MHA/GQA/Hybrid/MLA KV-cache scaling from 2K to 128K context, and self-hosted vs token API break-even, (2) Two-Stage Retrieval & Reranking (https://largo.dev/lab/?tool=retrieval) modeling BM25 + dense RRF fusion and cross-encoder latency SLOs, and (3) Stateful Agent Context & Memory (https://largo.dev/lab/?tool=memory) comparing quadratic transcript growth against active-slot consolidation. - [Companion Code Repository](https://github.com/steve-largo/largo-tutorials): 10 runnable Python packages with Pydantic v2 contracts, offline pytest suites, and reproducible benchmark CLIs across AI agents, embeddings, retrieval systems, production ML serving, and predictive maintenance. - [Articles & Concept Constellation](https://largo.dev/articles/): Technical essays and interactive knowledge graph connecting shared engineering concepts. - [Practitioner's Reference Shelf](https://largo.dev/resources/): Foundational papers, production runtimes, and evaluation datasets. - [About Steven W. White](https://largo.dev/about/): Author background, engineering principles, and side projects. - [RSS Feed](https://largo.dev/rss.xml): Chronological XML feed of all tutorials and technical analyses. - [XML Sitemap](https://largo.dev/sitemap-index.xml): Canonical URL index with last-modified timestamps. ## Technical Essays & Architecture Analyses - [Where Production AI Reliability Comes From: State, Retrieval, Memory, and Serving Economics](https://largo.dev/articles/anatomy-of-production-ai-2026/): A synthesis of measured benchmarks across largo.dev: why typed state graphs beat unconstrained prompt loops, how hybrid RRF and cross-encoders fix vector blind spots, why slot consolidation resolves mutating memory, and how KV-cache math governs GPU break-even. (2026-11-12, 17 min read) - [What It Takes to Be a Senior Machine Learning Engineer](https://largo.dev/articles/senior-mle-guide/): A field guide to the eight systems pillars behind production ML and AI engineering: GPU sizing, experiment lineage, CI/CD, serving compilation, two-stage hybrid retrieval, stateful agents, drift monitoring, and security. (2026-01-24, updated 2026-11-12, 8 min read) - [2026 Frontier LLM Architectures: MLA, iRoPE, Hybrid Sliding Attention, and mHC](https://largo.dev/articles/frontier-llm-architectures-2026/): Technical comparison of DeepSeek V3.2, Gemma 4 26B-A4B MoE, Llama 4, Gemini 3, and Qwen3: KV-cache bytes-per-token math at 128K context, Per-Layer Embeddings, and Manifold-Constrained Hyper-Connections. (2026-01-03, updated 2026-11-12, 11 min read) - [2025: The Year AI Got a Reality Check](https://largo.dev/articles/2025-year-in-review/): From DeepSeek R1's GRPO post-training efficiency to typed agent state graphs and 80% SWE-bench coding models, here is what materially changed for AI practitioners in 2025. (2025-12-31, updated 2026-11-12, 9 min read) - [2024: The Year AI Got Practical](https://largo.dev/articles/2024-year-in-review/): An engineering retrospective on the 2024 shifts that held up in production: repository-scale debugging, quantized open-weight GQA models, two-stage hybrid retrieval, and grammar-constrained tool calling. (2024-12-30, updated 2026-11-12, 9 min read) ## Tutorials: AI Agents, Stateful Memory & Open-Weight Serving - [Open-Weight Agent Serving: LoRA-Tuning Gemma 4 26B-A4B MoE on Vertex AI with vLLM](https://largo.dev/tutorials/agents/gemma-4-lora-vllm-vertex-ai/): Size Gemma 4 26B-A4B MoE memory and hybrid KV caches, scope LoRA adapters around multimodal vision layers, and connect Google ADK 2.14 directly to vLLM on Vertex AI. (2026-11-11, 17 min read, depth: advanced | Code: https://github.com/steve-largo/largo-tutorials/tree/main/agents/gemma4-vertex-vllm) - [Stateful Agent Memory: Rolling Sessions vs Vector RAG vs Vertex AI Memory Bank](https://largo.dev/tutorials/agents/stateful-agent-memory-vertex-ai/): Benchmark rolling ADK sessions, keyword recall, SQLite vector RAG, and consolidated slot memory across an 8-session coastal history using Google ADK 2.14 and Vertex AI Memory Bank. (2026-11-11, 15 min read, depth: advanced | Code: https://github.com/steve-largo/largo-tutorials/tree/main/agents/agent-memory-benchmark) - [Agent Orchestration in Practice: Google ADK vs LangGraph vs AWS Strands](https://largo.dev/tutorials/agents/adk-vs-langgraph-vs-strands/): Build the same live NOAA, Open-Meteo, and NWS marine report workflow in Google ADK 2.x and LangGraph on Vertex AI, then compare state graphs, parallel fan-out, failure handling, and LLM round-trip costs against AWS Strands. (2026-11-11, 15 min read, depth: advanced | Code: https://github.com/steve-largo/largo-tutorials/tree/main/agents/adk-vs-langgraph) - [Building an AI Trading Agent with Claude and News Signals](https://largo.dev/tutorials/agents/trading-agent/): Build an automated trading agent that extracts market signals from news using Claude Haiku, executes trades via Alpaca, and manages positions with trailing stops and sentiment monitoring. (2026-01-17, 11 min read, depth: advanced) - [Build a Community Christmas Tree with AI-Generated Ornaments](https://largo.dev/tutorials/agents/christmas-tree-ornament-generator/): Create a shared Christmas tree where visitors add AI-generated ornaments using Amazon Nova Canvas, with defense-in-depth content moderation using Bedrock Guardrails and Claude. (2025-12-22, 9 min read, depth: intermediate) - [Build a Holiday Cocktail Agent with TheCocktailDB](https://largo.dev/tutorials/agents/holiday-cocktail-agent/): Create an AI bartender that suggests cocktails based on weather, searches by ingredient, and generates party menus with shopping lists. (2025-12-22, 7 min read, depth: intermediate) - [Building a Fishing Report Agent with AWS Strands](https://largo.dev/tutorials/agents/fishing-report-agent/): Create an AI agent that combines tide, weather, and marine data to generate fishing reports. Learn tool-calling patterns with the Strands SDK, NOAA APIs, and Claude on AWS Bedrock. (2025-12-21, 13 min read, depth: intermediate) ## Tutorials: Production ML Infrastructure - [ML Security Best Practices](https://largo.dev/tutorials/production-ml/ml-security/): Secure your ML infrastructure with IAM roles, secrets management, VPC configuration, and input validation. Practical patterns for production systems. (2026-01-24, 8 min read, depth: advanced) - [ML Monitoring and Drift Detection](https://largo.dev/tutorials/production-ml/ml-monitoring/): Monitor production ML models with data drift detection, performance tracking, and automated alerting. Includes working Python implementations. (2026-01-24, 8 min read, depth: advanced) - [Model Serving on AWS](https://largo.dev/tutorials/production-ml/model-serving/): Deploy ML models to production with optimized inference: torch.compile vs ONNX benchmarks, FastAPI serving patterns, and AWS deployment options. (2026-01-24, 8 min read, depth: advanced | Code: https://github.com/steve-largo/largo-tutorials/tree/main/production-ml/model-serving) - [CI/CD for Machine Learning](https://largo.dev/tutorials/production-ml/ml-cicd/): Build a complete ML pipeline with GitHub Actions: data validation, model training, automated testing, and staged deployment to production. (2026-01-24, 10 min read, depth: advanced) - [Experiment Tracking with MLflow and Langfuse](https://largo.dev/tutorials/production-ml/experiment-tracking/): Set up experiment tracking for ML models with MLflow and LLM observability with Langfuse. Includes hyperparameter sweeps, model registry, and cost tracking. (2026-01-24, 11 min read, depth: intermediate) - [GPU Sizing for ML Workloads](https://largo.dev/tutorials/production-ml/gpu-sizing-for-ml/): Learn to calculate VRAM requirements, select the right AWS instance, and optimize costs. Includes real benchmarks and a Python sizing calculator. (2026-01-24, 12 min read, depth: intermediate) - [Data Models for AI Applications: Pydantic vs Python Built-ins](https://largo.dev/tutorials/production-ml/data-models-for-ai-applications/): Compare Python's data modeling options for AI/ML applications. Learn when to use dataclasses, TypedDict, or Pydantic for API responses, embeddings metadata, and agent tool contracts. (2026-01-01, 9 min read, depth: intermediate) - [Mamba for Predictive Maintenance: State Space Models vs Transformers](https://largo.dev/tutorials/production-ml/mamba-predictive-maintenance/): Compare Mamba's selective state space architecture against LSTM and Transformer for hard drive failure prediction. Learn when SSMs beat attention. (2025-12-29, 10 min read, depth: advanced | Code: https://github.com/steve-largo/largo-tutorials/tree/main/predictive-maintenance/hard-drive-failure) - [Predicting Hard Drive Failures with XGBoost, LSTM, and Transformers](https://largo.dev/tutorials/production-ml/hard-drive-failure-prediction/): Build a production-ready failure prediction system using real Backblaze data. Compare traditional ML vs deep learning approaches and learn when each shines. (2025-12-20, 18 min read, depth: intermediate | Code: https://github.com/steve-largo/largo-tutorials/tree/main/predictive-maintenance/hard-drive-failure) ## Tutorials: Embeddings & Reranking - [Cross-Attention Fusion: Combining Text Embeddings with Structured Features](https://largo.dev/tutorials/embeddings/cross-attention-text-tabular-fusion/): Concatenation is the default. Here's why cross-attention works better for combining text embeddings with tabular data, and how to implement it in PyTorch. (2026-01-14, 18 min read, depth: advanced) - [Cross-Encoders: Precision Reranking for Search](https://largo.dev/tutorials/embeddings/cross-encoders-reranking/): When bi-encoders aren't accurate enough, cross-encoders dramatically improve search relevance. Build a two-stage retrieval system with MS MARCO rerankers and sentence-transformers. (2025-12-21, 9 min read, depth: intermediate | Code: https://github.com/steve-largo/largo-tutorials/tree/main/embeddings/cross-encoders-reranking) - [Bi-Encoders: Fast Semantic Search at Scale](https://largo.dev/tutorials/embeddings/bi-encoders-semantic-search/): Learn how bi-encoders enable sub-millisecond semantic search over millions of documents. Build a complete search system with sentence-transformers, FAISS indexing, and production-ready Python code. (2025-12-21, 9 min read, depth: intermediate | Code: https://github.com/steve-largo/largo-tutorials/tree/main/embeddings/bi-encoders-semantic-search) - [Visual Semantic Search with CLIP and AWS](https://largo.dev/tutorials/embeddings/visual-semantic-search/): Build a production image search system using OpenAI's CLIP model, Amazon OpenSearch Serverless for vector storage, and Claude on Bedrock for image descriptions. Complete Python implementation with real AWS outputs. (2025-12-21, 16 min read, depth: advanced) - [Sentence Embeddings from Scratch with PyTorch](https://largo.dev/tutorials/embeddings/sentence-embeddings-from-scratch/): Build a complete sentence encoder from the ground up. Learn tokenization, embedding layers, pooling strategies, and benchmark on semantic similarity. (2025-12-20, 7 min read, depth: beginner | Code: https://github.com/steve-largo/largo-tutorials/tree/main/embeddings/sentence-embeddings-from-scratch) ## Tutorials: Transformers & Reasoning Architectures - [DeepSeek V3.2: Frontier Reasoning at 6x Lower Cost](https://largo.dev/tutorials/transformers/deepseek-v3-architecture/): Technical deep dive into DeepSeek V3.2's architecture: DeepSeek Sparse Attention (DSA), integrated reasoning with tool-use, and how it achieves IMO gold-medal performance. (2026-01-03, 11 min read, depth: advanced) - [DeepSeek R1 Architecture: How Reinforcement Learning Creates Reasoning](https://largo.dev/tutorials/transformers/deepseek-r1-architecture/): Deep dive into DeepSeek R1's architecture: how pure RL training enables chain-of-thought reasoning, the GRPO algorithm, MoE design, and knowledge distillation. (2025-02-01, 10 min read, depth: advanced) - [Getting Started with DeepSeek R1: The Open-Source Reasoning Model](https://largo.dev/tutorials/transformers/deepseek-r1-getting-started/): Run DeepSeek R1 on AWS Bedrock or locally with Ollama. Learn to use its chain-of-thought reasoning for complex problem-solving, coding, and math. (2025-02-01, 10 min read, depth: intermediate) ## Tutorials: Retrieval Systems & Vector Indexing - [Two-Stage Hybrid Retrieval: BM25, Dense Embeddings, RRF, and Cross-Encoder Reranking](https://largo.dev/tutorials/retrieval-systems/hybrid-search-reranking/): Build and benchmark a two-stage hybrid search pipeline combining Okapi BM25, BGE bi-encoder embeddings, Reciprocal Rank Fusion (RRF), and MS MARCO cross-encoder reranking across exact-identifier and semantic-paraphrase slices. (2026-11-12, 22 min read, depth: intermediate | Code: https://github.com/steve-largo/largo-tutorials/tree/main/retrieval-systems/hybrid-search-reranking) - [CFP Oracle: Semantic Search for College Football History](https://largo.dev/tutorials/retrieval-systems/cfp-oracle/): Build a semantic search system to find historically similar College Football Playoff games using Amazon S3 Vectors and Bedrock embeddings. (2026-01-01, 5 min read, depth: intermediate | Code: https://github.com/steve-largo/largo-tutorials/tree/main/retrieval-systems/cfp-oracle) - [Getting Started with Amazon S3 Vectors](https://largo.dev/tutorials/retrieval-systems/s3-vectors-getting-started/): Build a semantic search system using AWS's new serverless vector storage. Store millions of embeddings in S3 with sub-second query times and serverless pricing. (2025-12-31, 8 min read, depth: intermediate)