TheAILAB Blog https://theailab.party/blog/ Articles about automating business work, running AI on your own hardware and automating trading strategies. en-au Sun, 18 Oct 2026 07:20:00 GMT Build vs buy for internal tools: a practical decision framework https://theailab.party/blog/build-vs-buy-internal-tools-decision-framework/ https://theailab.party/blog/build-vs-buy-internal-tools-decision-framework/ Sun, 18 Oct 2026 07:20:00 GMT Custom Software A short set of questions for deciding whether to build, buy or configure an internal tool, drawn from government guidance and engineering practice. Data leakage through AI assistants: where your pasted text goes https://theailab.party/blog/data-leakage-through-ai-assistants/ https://theailab.party/blog/data-leakage-through-ai-assistants/ Sat, 17 Oct 2026 21:20:00 GMT Data Privacy & Security What OpenAI, Anthropic and Google say happens to the text you paste into ChatGPT, Claude and Gemini: training, human review, retention and business plans. Measuring hallucinations in document Q&A: methods that hold up https://theailab.party/blog/measuring-hallucinations-in-document-qa/ https://theailab.party/blog/measuring-hallucinations-in-document-qa/ Sat, 17 Oct 2026 07:20:00 GMT AI Automation How to measure hallucinations in a document Q&A or RAG system: claim-level faithfulness, citation checks, published benchmarks and hand grading. Human-in-the-loop design: where approval steps belong in AI workflows https://theailab.party/blog/human-in-the-loop-approval-steps-ai-workflows/ https://theailab.party/blog/human-in-the-loop-approval-steps-ai-workflows/ Fri, 16 Oct 2026 21:30:00 GMT AI Automation Where to put human approval in an AI workflow, what the reviewer should see, and how to stop approvals turning into a rubber stamp. Local RAG: running retrieval-augmented generation on your own machines https://theailab.party/blog/local-rag-on-your-own-machines/ https://theailab.party/blog/local-rag-on-your-own-machines/ Fri, 16 Oct 2026 07:30:00 GMT Local AI How to build retrieval-augmented generation that never leaves your hardware: embeddings, storage, hybrid search, prompting and the security gaps to close. LLM quantisation explained: what 4-bit and 8-bit models trade away https://theailab.party/blog/llm-quantisation-4-bit-vs-8-bit-trade-offs/ https://theailab.party/blog/llm-quantisation-4-bit-vs-8-bit-trade-offs/ Thu, 15 Oct 2026 21:45:00 GMT Local AI What 4-bit and 8-bit LLM quantisation costs in quality, why benchmark averages hide it, and how to test a quantised model on your own tasks. When not to use an LLM: tasks where plain code and rules win https://theailab.party/blog/when-not-to-use-an-llm/ https://theailab.party/blog/when-not-to-use-an-llm/ Thu, 15 Oct 2026 07:26:00 GMT AI Automation How to tell which steps in a workflow need a language model and which are better as plain code or rules, with a simple test and a hybrid pattern. Least-privilege API keys and secrets for your automations https://theailab.party/blog/least-privilege-api-keys-secrets-management-automations/ https://theailab.party/blog/least-privilege-api-keys-secrets-management-automations/ Wed, 14 Oct 2026 21:20:00 GMT Data Privacy & Security How to scope, store, rotate and revoke the API keys your scripts, workflows and AI agents use, based on OWASP, GitHub, AWS and Google Cloud guidance. Position sizing basics: fixed fractional, vol targeting and Kelly https://theailab.party/blog/position-sizing-fixed-fractional-volatility-targeting-kelly/ https://theailab.party/blog/position-sizing-fixed-fractional-volatility-targeting-kelly/ Wed, 14 Oct 2026 07:20:00 GMT Algorithmic Trading How fixed fractional sizing, volatility targeting and the Kelly criterion decide trade size, with worked numbers and the caveats that matter in live trading. Open-weight model licences: what you can do commercially https://theailab.party/blog/open-weight-model-licences-commercial-use/ https://theailab.party/blog/open-weight-model-licences-commercial-use/ Tue, 13 Oct 2026 21:35:00 GMT Local AI What the Llama 4, Gemma, Apache 2.0 and Mistral licence texts actually allow for commercial use, fine-tuning, redistribution and naming. NIST's AI risk management framework, explained in plain English https://theailab.party/blog/nist-ai-risk-management-framework-plain-english/ https://theailab.party/blog/nist-ai-risk-management-framework-plain-english/ Tue, 13 Oct 2026 07:32:00 GMT Data Privacy & Security What the NIST AI RMF actually asks you to do, how Govern, Map, Measure and Manage fit together, and how a small team can use it without drowning in paperwork. Structured outputs and function calling for reliable LLM automation https://theailab.party/blog/structured-outputs-and-function-calling-for-llm-automation/ https://theailab.party/blog/structured-outputs-and-function-calling-for-llm-automation/ Mon, 12 Oct 2026 21:25:00 GMT AI Automation What structured outputs and function calling guarantee, where the guarantee stops, and how to design schemas and checks so an LLM can safely feed an automation. SQLite in production: when it's the right choice and when it isn't https://theailab.party/blog/sqlite-in-production-when-its-the-right-choice/ https://theailab.party/blog/sqlite-in-production-when-its-the-right-choice/ Mon, 12 Oct 2026 07:31:00 GMT Custom Software Use the SQLite docs' own checklist to decide if it suits your internal tool, then set up WAL, busy timeouts, foreign keys and safe backups. The Australian Privacy Act and AI tools: a guide for small businesses https://theailab.party/blog/australian-privacy-act-ai-tools-small-business/ https://theailab.party/blog/australian-privacy-act-ai-tools-small-business/ Sun, 11 Oct 2026 21:20:00 GMT Data Privacy & Security What the OAIC expects when your business uses ChatGPT-style tools or automated decisions, and the new privacy policy rule that starts on 10 December 2026. How to evaluate an AI automation before you trust it https://theailab.party/blog/how-to-evaluate-an-ai-automation/ https://theailab.party/blog/how-to-evaluate-an-ai-automation/ Sun, 11 Oct 2026 07:30:00 GMT AI Automation Build a test set from real work, grade outputs with code and careful LLM judges, read the failures, and decide where human review stays before you go live. Ollama vs llama.cpp vs vLLM: which local LLM tool fits the job https://theailab.party/blog/ollama-vs-llama-cpp-vs-vllm/ https://theailab.party/blog/ollama-vs-llama-cpp-vs-vllm/ Sat, 10 Oct 2026 21:25:00 GMT Local AI How Ollama, llama.cpp and vLLM differ in hardware support, concurrency and setup, and how to pick one for a laptop, a team server or a busy API. How much VRAM do you need to run an LLM locally? https://theailab.party/blog/how-much-vram-to-run-an-llm-locally/ https://theailab.party/blog/how-much-vram-to-run-an-llm-locally/ Sat, 10 Oct 2026 18:20:00 GMT Local AI Work out the GPU memory a local LLM needs from its parameter count, quantisation and context length, with worked examples from official docs. Why your backtest looks better than your live trading results https://theailab.party/blog/why-your-backtest-looks-better-than-live-results/ https://theailab.party/blog/why-your-backtest-looks-better-than-live-results/ Sat, 10 Oct 2026 16:00:00 GMT Algorithmic Trading Overfitting, look-ahead bias, survivorship bias and missing costs explain why most backtests disappoint in live trading. Here's how to catch each one. Prompt injection: the security risk every AI automation must plan for https://theailab.party/blog/prompt-injection-ai-automation-security/ https://theailab.party/blog/prompt-injection-ai-automation-security/ Sat, 10 Oct 2026 12:00:00 GMT Data Privacy & Security Prompt injection lets text hidden in emails, documents or web pages hijack an AI assistant. What it is, why there's no full fix, and how to design around it.