Start

Blog

LatestOctober 7, 202626 min readOptimizing LLM AppsPrompt engineering, retrieval, and fine-tuning: when each one pays off, and the order to try them in.Stephen M. Walker IICo-Founder, CEO
26 posts

2026

Jul 14Fine-tuningFine-Tuning Guide: When It Is Worth It and How to Do It in 2026A decision framework and production workflow for fine-tuning LLMs, from prompting and RAG through datasets, evaluation, training, deployment, and monitoring.Jul 14Fine-tuningFine-Tuning OpenAI Models: What Still WorksThe OpenAI models you can still fine-tune, the current API syntax and pricing, and what to decide before self-serve fine-tuning closes.Jul 14Fine-tuningFine-Tuning GPT-4 with OpenAI's API: A Current GuideWhat remains fine-tunable in the GPT-4 family, the current API and data format, and the available alternatives.Jul 14EvaluationHow Teams Actually Agree a Prompt Got BetterA repeatable workflow for mixed teams to decide whether a new prompt version is actually better, and to keep a written record of why.Jul 14EngineeringThe Best AI Tools for Startups in 2026: A Stack That Earns Its CostAn AI tool stack for startups sorted by job, with what each tool sees about your company and when to skip it.

2025

Dec 15EvaluationComparative Analysis: Google Gemini Pro vs. OpenAI GPT-3.5Gemini Pro beat GPT-3.5 on speed, reading comprehension, instruction following, and code, but its API design and safety blocking kept us on OpenAI.Dec 8EngineeringBuilding Your AI Team in 2025Five team shapes, the capabilities every team needs, and a working evaluation loop, from a year of conversations with AI product teams.Dec 8EngineeringPrompt Engineering Guide: A Comprehensive Examination of Prompt TechniquesPrompt formats for each major model family, techniques like few-shot and chain-of-thought, and how to evaluate and iterate on prompts.Nov 23IndustryAnalyzing Retool's State of AI 2023 ReportKey findings from a survey of over 1,500 tech professionals, and what they mean for teams shipping AI features.Nov 10IndustryOpenAI DevDay 2023 ReflectionsThe Klu team evaluates the features OpenAI launched at DevDay 2023.Oct 19GuidesStartup Guide to Azure OpenAIHow Azure OpenAI differs from the public API, and how to use its speed and startup credits.Oct 16IndustryAI Safety in 2023: analysts claim issues up 700% in 2023What the numbers say about AI safety in 2023, the year generative models dominated headlines and sparked calls for regulation.Oct 10EngineeringWhy most Vector DBs will die – Pt.01Why Klu built its vector search on pgvector and Supabase instead of Pinecone, and why retrieval latency barely registers next to LLM generation time.Oct 6IndustryAI in the News: Alignment, Doom, Risk, and UnderstandingCommon misconceptions about understanding, controlling, and aligning AI systems, broken down.Sep 23GuidesGuide: Getting started with Klu Python SDKBuild your first Klu Action with the Python SDK in five steps.Sep 23GuidesGuide: Getting started with Klu TypeScript SDKBuild your first Klu Action with the TypeScript SDK in five steps.Sep 22EngineeringJust Launched: Our expanded, global OpenAI GPT-4 deploymentKlu's Azure OpenAI GPT-4 deployment across 15 regions, the full region list, and why redundancy, load scaling, and regional privacy drove it.Sep 1EvaluationEverything We Know About GPT-4A GPT-4 model card compiled from public research: about 1.8 trillion parameters, mixture-of-experts architecture, training compute, data, and known limitations.Sep 1Fine-tuningOpenAI SDK: Fine-tune GPT-3.5 TurboFormat, clean, and upload your data, then train and evaluate a GPT-3.5 Turbo fine-tune, with a Joe Rogan comedy model as the example.Aug 28EvaluationOpen Source LLMs: Exploring 5 Popular Models in August 2023Llama 2, Vicuna, Falcon, MPT, and StableLM compared on architecture, training data, and benchmarks, plus three ways to run them on a laptop.Aug 19IndustryAI Collaboration: the key to breakout productsHow companies pair human judgment with AI in healthcare, marketing, and manufacturing, and the ethics and reskilling work that adoption requires.Jul 17EvaluationBest Open Source LLMs of 2025A ranked list of open source LLMs, plus how to test them online, run them locally, fine-tune them, and serve them in production.Jul 12EvaluationEvaluating 2024 Frontier Model Capabilities Pt.01We built QUAKE to test frontier models on everyday knowledge-work tasks, and the 2024 models averaged 28% under human review.Jul 8EvaluationUnleashing the Power of Multimodal AI Models: Understanding the Future of Artificial IntelligenceHow GPT-4V, Gemini Vision, LLaVA 1.5, and Fuyu-8B handle text and images, with a leaderboard, integration challenges, and real applications.Jul 4IndustryAI Readiness: 7 Key Areas to Evaluate for Successful AI ImplementationSeven areas to assess before adopting AI: strategy, culture, data, infrastructure, ethics, change management, and how you measure success.