Blog
Latest articles
7 min readdbt: SQL-Based Data Transformation That Brings Software Engineering to Analytics
dbt (data build tool) lets analysts transform data in the warehouse using SELECT statements, with built-in testing, documentation, and dependency tracking - no more unmaintainable SQL scripts.
9 min readDPO: The RLHF Alternative That Trains LLMs Without a Reward Model
Direct Preference Optimization eliminates the separate reward model and PPO loop from RLHF, deriving an equivalent alignment objective directly from preference data with a simple classification loss.
7 min readInternVL 2: The Open-Source VLM Approaching GPT-4V on Benchmarks
Shanghai AI Lab's InternVL2-26B scores 61.2% on MMMU - within 2 points of GPT-4V - using a 6B vision encoder and dynamic high-resolution image tiling.
7 min readTailwind CSS v4: CSS-First Configuration and 5x Faster Builds
Tailwind CSS v4 replaces tailwind.config.js with a CSS-first `@theme` directive and swaps PostCSS for Lightning CSS - here is everything that changed and how to migrate.
8 min readGoogle AI Overviews: How to Appear in AI-Generated Summaries
Google AI Overviews are reshaping search CTR patterns. Here is what triggers them, how Google selects sources, and how to optimize your content to be cited.
7 min readDistilBERT in Production: Fast NLP Classification Without the GPU Bill
DistilBERT delivers 97% of BERT's performance at 40% smaller size and 60% faster inference, making it the practical default for production text classification that needs low latency on CPU.
7 min readRunway Gen-3 Alpha: Professional Video Generation for AI Filmmakers
Runway Gen-3 Alpha raises the bar for AI video with dramatically improved temporal consistency, a motion brush tool for selective animation, and Act One for transferring facial performances onto generated characters.
7 min readMistral Large 2: The 128K Context Enterprise LLM from Europe
Mistral Large 2 packs 123B parameters, 128k context, and support for 80+ languages into a model that scores 92% on HumanEval and costs $2/1M input tokens.
8 min readSemantic Kernel: Microsoft's SDK for Building AI Apps in C# and Python
Semantic Kernel gives .NET and Python developers a structured way to combine LLMs, plugins, memory, and multi-step planners into production AI applications.
9 min readSpeculative Decoding: How to Get 3x LLM Speed With a Smaller Draft Model
Speculative decoding uses a small fast model to draft multiple tokens and a large model to verify them in parallel, achieving 1.5-3x speedups without changing output distribution.
8 min readLangfuse: Open-Source LLM Observability You Can Self-Host
Langfuse brings full tracing, prompt versioning, dataset evaluation, and cost attribution to LLM apps - and you can run the entire stack on your own servers.
8 min readSchema Markup for Rich Results: The 8 Types That Drive the Most Traffic
JSON-LD structured data can get your content into Google's rich result formats - star ratings, FAQ dropdowns, how-to steps, and more. Here are the eight types worth implementing.