https://www.nedved.tech/blog Field notes from real AI-first delivery, multi-agent systems, LLM deep dives, and transformation thinking. en-sg Fri, 25 Sep 2026 06:53:51 GMT https://www.nedved.tech/blog/beyond-code-ai-dlc https://www.nedved.tech/blog/beyond-code-ai-dlc Thu, 03 Sep 2026 00:00:00 GMT As the AI development lifecycle matures, code stops being the only first-class artefact — and one delivery pipeline stops being enough. The latest chapter in the AI-DLC series. https://www.nedved.tech/blog/geek-out-time-agent-harness-exploration-part-1-what-happens-when-success-is-impossible https://www.nedved.tech/blog/geek-out-time-agent-harness-exploration-part-1-what-happens-when-success-is-impossible Tue, 21 Jul 2026 00:00:00 GMT I gave my coding agent an impossible task. It eventually redefined “equals.” https://www.nedved.tech/blog/ai-coding-made-us-faster https://www.nedved.tech/blog/ai-coding-made-us-faster Mon, 06 Jul 2026 00:00:00 GMT Reflections from a first real AI-native project built with Claude Code and Codex: when coding becomes fast, the bottleneck moves to requirements, testing, UAT, and ownership. https://www.nedved.tech/blog/from-ai-coding-to-ai-sdlc https://www.nedved.tech/blog/from-ai-coding-to-ai-sdlc Sat, 16 May 2026 00:00:00 GMT Four iterations into a real AI-first project: why AI agents are like very capable (but imperfect) developers, and why traditional engineering discipline matters more, not less. https://www.nedved.tech/blog/geek-out-time-can-ai-do-research-by-itself-i-tried-autoresearch-on-singapore-hawker-food https://www.nedved.tech/blog/geek-out-time-can-ai-do-research-by-itself-i-tried-autoresearch-on-singapore-hawker-food Sun, 05 Apr 2026 00:00:00 GMT Andrej Karpathy’s autoresearch ( https://github.com/karpathy/autoresearch ) hit 62k GitHub stars in weeks. The idea is genuinely interesting: give an AI agent a real LLM training setup, let it run experiments autonomously overnight, wake up to a log of 100 experiments and hopefully a better model. T https://www.nedved.tech/blog/fewer-developers-more-builders-part-2 https://www.nedved.tech/blog/fewer-developers-more-builders-part-2 Fri, 27 Feb 2026 00:00:00 GMT Week one of a deliberately AI-first project: same AI, different people, different results — why AI-first development is a capability shift, not a tooling shift. https://www.nedved.tech/blog/fewer-developers-more-builders https://www.nedved.tech/blog/fewer-developers-more-builders Mon, 23 Feb 2026 00:00:00 GMT Why we committed a real internal product to a deliberately AI-first SDLC — and why the future needs fewer traditional developers but many more builders. https://www.nedved.tech/blog/geek-out-time-gradient-boosted-trees-gbt-using-a-hallucination-risk-playground https://www.nedved.tech/blog/geek-out-time-gradient-boosted-trees-gbt-using-a-hallucination-risk-playground Thu, 19 Feb 2026 00:00:00 GMT Gradient Boosted Trees (GBT) are one of those models everyone uses, and u probably know the story: sequential trees, correcting residuals, learning rate, depth trade-offs. This week’s geek-out was about playing with GBT in a modern setting — a synthetic “LLM hallucination risk” dataset. Not because https://www.nedved.tech/blog/geek-out-time-tree-of-thought-for-llm-reasoning https://www.nedved.tech/blog/geek-out-time-tree-of-thought-for-llm-reasoning Wed, 11 Feb 2026 00:00:00 GMT As LLMs get better at producing structured explanations, it has become increasingly common to describe them as “reasoning models.” The release of OpenAI’s o-series, particularly o1, along with DeepSeek R1 etc, made this framing much more explicit. Unlike earlier models that relied primarily on scali https://www.nedved.tech/blog/geek-out-time-running-fastmcp-with-deepseek-on-google-colab https://www.nedved.tech/blog/geek-out-time-running-fastmcp-with-deepseek-on-google-colab Sun, 25 Jan 2026 00:00:00 GMT I’ve been reading quite a bit about the MCP recently. It is an open standard designed to solve the “integration tax” of LLMs. Instead of writing customized glue code for every tool or data source, MCP provides a universal interface. Think of it like a Language Server Protocol (LSP) but for AI: it st https://www.nedved.tech/blog/geek-out-time-reject-by-design-determinism-in-enterprise-ai https://www.nedved.tech/blog/geek-out-time-reject-by-design-determinism-in-enterprise-ai Fri, 09 Jan 2026 00:00:00 GMT In the previous geek out, we established a boundary that enterprise AI systems cannot afford to blur: language understanding is not the same thing as authority. Ontology gives us a way to model reality explicitly, and a deterministic core gives us something regulators can actually trust. https://www.nedved.tech/blog/geek-out-time-from-ontology-to-ai-agents-determinism-matters-in-high-compliance-industries https://www.nedved.tech/blog/geek-out-time-from-ontology-to-ai-agents-determinism-matters-in-high-compliance-industries Mon, 05 Jan 2026 00:00:00 GMT In my earlier geek out, “Why Enterprise AI Quietly Depends on Ontology”, we explored a simple but easily overlooked idea: enterprise AI doesn’t fail because models aren’t smart enough. It fails because reality isn’t modeled explicitly enough. https://www.nedved.tech/blog/geek-out-time-why-rag-alone-fails-enterprise-governance-and-how-ontology-fixes-it https://www.nedved.tech/blog/geek-out-time-why-rag-alone-fails-enterprise-governance-and-how-ontology-fixes-it Wed, 31 Dec 2025 00:00:00 GMT In the earlier geek out ( https://medium.com/the-constellar-digital-technology-blog/geek-out-why-enterprise-ai-quietly-depends-on-ontology-485bdecc733c ), I focused on the separation of responsibilities: letting LLMs deal with language, while ontology enforces rules. That experiment assumed somethin https://www.nedved.tech/blog/geek-out-why-enterprise-ai-quietly-depends-on-ontology https://www.nedved.tech/blog/geek-out-why-enterprise-ai-quietly-depends-on-ontology Fri, 19 Dec 2025 00:00:00 GMT Ontology is one of those ideas that never really went away — it just stayed invisible. Long before LLM, ontology was already being used to model how complex systems work: what entities exist, how they relate, and which actions are allowed. If u are a Palantir stock holders, u won’t feel surprised th https://www.nedved.tech/blog/geek-out-time-can-a-model-detect-when-its-about-to-hallucinate https://www.nedved.tech/blog/geek-out-time-can-a-model-detect-when-its-about-to-hallucinate Sun, 14 Dec 2025 00:00:00 GMT Hallucinations are usually discussed after the fact. The answer is wrong. The reference doesn’t exist. A confident explanation collapses the moment a human checks it. At that point, we talk about better prompts, stricter retrieval, guardrails, or simply using a bigger model. Before a model hallucina https://www.nedved.tech/blog/geek-out-time-the-economics-of-llms-how-token-pricing-quietly-shapes-the-architecture https://www.nedved.tech/blog/geek-out-time-the-economics-of-llms-how-token-pricing-quietly-shapes-the-architecture Sun, 07 Dec 2025 00:00:00 GMT Every time a new LLM comes out, we tend to compare quality, benchmarks, reasoning scores. But when you are actually running something at scale — say a customer support agent handling tens of thousands of queries per month — the story becomes much less glamorous and much more economic. The architectu https://www.nedved.tech/blog/geek-out-time-testing-tpu-vs-gpu-on-google-colab-after-metas-reported-shift-toward-tpus https://www.nedved.tech/blog/geek-out-time-testing-tpu-vs-gpu-on-google-colab-after-metas-reported-shift-toward-tpus Wed, 26 Nov 2025 00:00:00 GMT The AI hardware discussion has heated up over the past two days after reports that Meta is exploring using Google TPUs for parts of its training workloads. The market responded immediately. Google traded higher, NVIDIA slipped, and the usual questions began circulating: Is NVIDIA’s dominance finally https://www.nedved.tech/blog/geek-out-time-where-llms-keep-their-facts-vscode-with-colab https://www.nedved.tech/blog/geek-out-time-where-llms-keep-their-facts-vscode-with-colab Mon, 17 Nov 2025 00:00:00 GMT A few weeks ago, we tried “ Reducing Hallucination with DoLa in Google Colab (Microsoft Phi-2)”( https:// nedvedyang.medium.com/geek-out-time-reducing-hallucination-with-dola-in-google-colab-microsoft-phi-2-ec878793df50 ) , a technique that improves factual accuracy by mixing signals from different https://www.nedved.tech/blog/icf-acc-exam-prep-reflections-and-lessons-learned https://www.nedved.tech/blog/icf-acc-exam-prep-reflections-and-lessons-learned Mon, 03 Nov 2025 00:00:00 GMT I recently passed the ICF Associate Certified Coach (ACC) exam. It was a meaningful milestone — not just because of the certification itself, but because of how much the preparation process revealed about my own mindset as a problem-solver learning to coach. https://www.nedved.tech/blog/geek-out-time-deepseek-powered-query-expansion-and-self-querying-in-advanced-rag https://www.nedved.tech/blog/geek-out-time-deepseek-powered-query-expansion-and-self-querying-in-advanced-rag Sat, 25 Oct 2025 00:00:00 GMT RAG has rapidly become the most widely used architecture for enterprise-grade LLM systems. Why? Because it grounds large language models in private, up-to-date knowledge bases, drastically reducing hallucinations and improving trustworthiness. If you’re building a reliable, production-ready AI appli https://www.nedved.tech/blog/geek-out-time-when-embeddings-hit-their-limit https://www.nedved.tech/blog/geek-out-time-when-embeddings-hit-their-limit Tue, 14 Oct 2025 00:00:00 GMT In information retrieval, dense embeddings have become the standard way to represent and retrieve text. With better models and larger dimensions, we expect better recall. https://www.nedved.tech/blog/geek-out-time-reducing-hallucination-with-dola-in-google-colab-microsoft-phi-2 https://www.nedved.tech/blog/geek-out-time-reducing-hallucination-with-dola-in-google-colab-microsoft-phi-2 Mon, 06 Oct 2025 00:00:00 GMT Is decoding just about the top layer — greedy, beam, temperature sampling, maybe top-p if you want variation? What if we could make a model reason better just by changing how we decode, not what we train? https://www.nedved.tech/blog/geek-out-time-the-rich-get-smarter-in-rag-continuing-from-embeddings https://www.nedved.tech/blog/geek-out-time-the-rich-get-smarter-in-rag-continuing-from-embeddings Tue, 23 Sep 2025 00:00:00 GMT Geek Out Time: The “Rich” Get Smarter in RAG -Continuing from “ Embeddings ” https://www.nedved.tech/blog/geek-out-time-back-to-when-i-first-asked-what-is-a-container-cgroups-and-namespaces-on-colab https://www.nedved.tech/blog/geek-out-time-back-to-when-i-first-asked-what-is-a-container-cgroups-and-namespaces-on-colab Fri, 12 Sep 2025 00:00:00 GMT I was reading this post the other day — Why I Ditched Docker for Podman (and You Should Too) — and it reminded me of the first time I learned about containers from this excellent YouTube talk nearly nine years ago. Time flies. https://www.nedved.tech/blog/geek-out-time-from-prompts-to-programs-how-open-source-dspy-deepseek-train-better-autogen https://www.nedved.tech/blog/geek-out-time-from-prompts-to-programs-how-open-source-dspy-deepseek-train-better-autogen Thu, 04 Sep 2025 00:00:00 GMT Geek Out Time: From Prompts to Programs -How Open-Source DSPy + DeepSeek Train Better Autogen Coaching Agents https://www.nedved.tech/blog/embeddings-continuing-from-chunk-happens-rag https://www.nedved.tech/blog/embeddings-continuing-from-chunk-happens-rag Mon, 01 Sep 2025 00:00:00 GMT In our last Geek Out Time we poked at chunking — the unglamorous decision that quietly controls what even has a chance to be retrieved. This time we zoom in on the sibling that decides whether those chunks are actually found: embeddings. If chunking is how knowledge gets in , embeddings are how it’s https://www.nedved.tech/blog/geek-out-time-lora-fine-tuning-new-gemma-3-270m-on-a-free-colab-gpu https://www.nedved.tech/blog/geek-out-time-lora-fine-tuning-new-gemma-3-270m-on-a-free-colab-gpu Sun, 24 Aug 2025 00:00:00 GMT The Gemma family is Google’s initiative to provide lightweight, open-weight models for developers and researchers. The name “Gemma” comes from the Latin for precious stone , and these models are built from the same research foundation as Google’s closed-source Gemini models. https://www.nedved.tech/blog/geek-out-time-building-a-kid-safe-ai-agent-with-multi-layer-protection-tf-idf-content https://www.nedved.tech/blog/geek-out-time-building-a-kid-safe-ai-agent-with-multi-layer-protection-tf-idf-content Mon, 11 Aug 2025 00:00:00 GMT Geek Out Time: Building a Kid-Safe AI Agent with Multi-Layer Protection (TF-IDF & Content Moderation) in Google Colab https://www.nedved.tech/blog/training-ai-agents-for-tool-use-with-reinforcement-learning-in-google https://www.nedved.tech/blog/training-ai-agents-for-tool-use-with-reinforcement-learning-in-google Sun, 03 Aug 2025 00:00:00 GMT R𝗚𝗲𝗲𝗸 𝗢𝘂𝘁 𝗧𝗶𝗺𝗲: Training AI Agents for Tool Use with Reinforcement Learning in Google Colab https://www.nedved.tech/blog/geek-out-time-human-in-the-loop-safety-for-high-stakes-ai-agents-humanlayer-gpt-4-on-google https://www.nedved.tech/blog/geek-out-time-human-in-the-loop-safety-for-high-stakes-ai-agents-humanlayer-gpt-4-on-google Thu, 31 Jul 2025 00:00:00 GMT Geek Out Time: Human-in-the-Loop Safety for High-Stakes AI Agents (HumanLayer + GPT-4 on Google Colab)