AI & Machine Learning
Building with AI models: the Gemini, OpenAI and Claude APIs, LLM integrations, industry playbooks, AI governance and the latest AI news.
Categories in AI & Machine Learning
Learning series
Articles in AI & Machine Learning (page 1 of 10)
- RAG Knowledge Base Freshness: Staleness and Incremental Updates — A RAG system is only as current as its index. This guide covers detecting changes, updating incrementally, handling deletions, measuring staleness and telling users honestly when an answer may be out of date.
- Re-Embedding Migrations: How to Change Embedding Models Safely — Vectors from different embedding models cannot be compared, so switching models means re-embedding everything. Here is a migration plan with versioned storage, backfill, evaluation before cutover, a rollback path and a cost estimate.
- pgvector HNSW vs IVFFlat: Index Tuning for RAG in Postgres — An exact vector scan works until it does not. Then you must pick an approximate index and tune it. This guide compares HNSW and IVFFlat in pgvector, explains the knobs that trade recall for speed, and shows how to measure the result on your own data.
- Multimodal RAG: Scanned PDFs, Charts and Page Images — Plain text extraction fails on scans, charts and complex tables, so RAG answers quietly go wrong. This guide compares OCR, layout parsing and vision models, explains page-image retrieval, and shows how to test the whole pipeline.
- Multilingual RAG: Cross-Language Retrieval and Answers — Your documents are in one language and your users write in another, or both are mixed. This guide compares multilingual embeddings with translation, shows how to evaluate each language separately, and how to control the answer language.
- Query Rewriting, Multi-Query and HyDE: Better Retrieval for RAG — Users ask short, vague or context-dependent questions that embed poorly. Rewriting the query before retrieval can help a lot or make things worse. Here is when to use each technique, what it costs, and how to measure the difference.
- Map-Reduce Summarization: Handling Long Documents in AI Automation — When a document does not fit in one call, or fits but performs badly, split, summarise the parts and combine them. This guide covers chunking choices, the combine step, faithfulness checks, cost, and failure handling in automation.
- Handling LLM Refusals and Content-Filter Blocks in Automation — Sometimes the model says no, or the API blocks the response, and a workflow that assumed a normal answer breaks in strange ways. Learn to detect each case, decide whether to retry, rephrase or escalate, and log it for review.
- Verifiable RAG: Citations, Grounding and Knowing When to Say "I Don't Know" — A RAG system that answers confidently but cannot show its work is a liability. Citation, grounding and abstention are what make answers checkable. Here is how to build each, verify citations in code and measure whether it is working.
- Context Engineering for AI Automation: What Actually Goes in the Prompt — Most prompt problems are context problems: too much, too little, or the wrong things in the wrong order. This guide covers what to include, how to budget and trim it, and how to test context changes like any other change.
- Chunking Strategies for RAG: How to Split Documents So Retrieval Actually Works — When a RAG system answers badly, the cause is often the step tutorials mention in one line: how the documents were cut into chunks. This guide compares the main strategies, handles tables, code and PDFs, and shows how to test chunking with evidence.
- Triaging Content Moderation Queues with AI Automation — A moderation queue sorted only by arrival time treats a mildly rude comment and a credible threat as equally urgent. AI automation's job is fixing that ordering, not making the final call on either one.
- Tracking Compliance Training Assignment and Completion with AI Automation — A spreadsheet tracking who completed mandatory training is fine until an auditor asks for proof three years later and the person who maintained it has left the company.
- Flagging Privileged and Responsive Documents in AI-Automated E-Discovery Review — A model that flags every document mentioning a lawyer as privileged either buries reviewers in false positives or, worse, misses the one email that should never have gone to opposing counsel.
- Routing Vendor Payment Approvals Through AI Automation Without Breaking Your Spend Matrix — An AI automation that reads a vendor payment request and decides who needs to approve it is only safe if it cannot quietly route around your actual delegation-of-authority rules.
- Flagging Risky Contract Clauses with AI Automation Before Signature — A legal team reading every incoming contract line by line does not scale. An AI automation that flags the specific clauses worth a lawyer's attention, and leaves the rest alone, does.
- The AI API Cache Pricing War: What Fable 5.1's Cut and Gemini's 2027 Price Hike Mean for Your Bill — Anthropic just cut cache pricing 75%, Google locked in a promotional rate that doubles on a fixed 2027 date, and xAI priced Grok 4.6 in the middle - here's what these real AI API cache pricing moves mean for your budget.
- DeepSeek V4.1 Flash: A 552B MoE Model Built to Cut Agent Memory Costs — DeepSeek V4.1 Flash is a 552-billion-parameter Mixture-of-Experts model released September 10, 2026, built around one specific goal: cutting the memory cost of running long-lived AI agents by roughly 4x.
- Decoding AI Model Names 2026: Fable, Mythos, Astra, Sol, Luna, Terra, and Muse Spark Explained — Fable, Mythos, Astra, Sol, Luna, Terra, Muse Spark. If you have lost track of which AI model is which, here is a plain-language guide to 2026's AI model names and the mental model for parsing the next one.
- Inside Trusted Access AI Model Programs: How Anthropic, OpenAI, and Google Gate Their Riskiest Models — Anthropic gates Claude Mythos 5.1 behind trusted-access programs for cybersecurity and life-sciences work, and it isn't alone. Here's what's confirmed about trusted access as a category, and what the cross-vendor pattern does and doesn't tell us.
- The AI Model Context Window Arms Race: What 1M+ Tokens Actually Buys You in 2026 — Gemini 3.8 Flash ships 1M tokens, Grok 4.6 ships 500K, Mistral Large 3 ships 256K. Here's what an AI model context window actually is, why bigger isn't automatically better, and when each size genuinely earns its cost.
- Open-Weight AI Models 2026 vs. Closed Frontier: Reading September's Ten-Day Release Sprint — Eleven models from seven providers shipped in a single ten-day window this September - and the order they landed in, open-weight AI models 2026 first and closed-frontier labs following within days, tells you more than any one benchmark score.
- Mistral Large 3, Small 4, and Medium 3.5: Mistral's Enterprise Open-Weight Lineup in 2026 — Mistral now ships a three-tier open-weight lineup, Mistral Large 3, Medium 3.5, and Small 4, each suited to a different mix of capability, latency, and cost. Here is how to pick the right tier.
- Meta Muse Spark 1.3 and the End of the Llama Name — Meta shipped Muse Spark 1.3 on September 2, 2026, continuing its shift away from the Llama name toward a new closed-weights model brand, even as older Llama models stay available.