nohe.dev

I like to tinker

Exploring the frontiers of AI, web development, and secure software architecture. Join me as I break things to see how they work.

Latest

View all posts
In this episode of Tech Sits, Rody and Nohe dive into their latest development projects and AI workflows. Topics covered in this episode: Genkit: An overview of the AI orchestration framework for full-stack applications. Continuous Improvement Loops (Kaizen): Using models like Mistral and Gemini to iteratively refine AI outputs. Vertex AI Model Garden: Accessing serverless models without managing infrastructure. Operational Transparency: Exploring how the Gemini CLI streams thinking tokens to improve user experience. Developer Portfolio Rebuilds: Rody shares his process for rebuilding his website using AI Studio, React, and Astro. Automating Metadata: Embedding Gemma 3N to auto-generate blog descriptions and surface related posts via KNN queries. Astro Galleries: Creating photo galleries with masonry layouts and lightboxes.
genkit • Mar 11, 2026

Genkit, AI Continuous Improvement & Astro Website Builds

In this episode of Tech Sits, Rody and Nohe dive into their latest development projects and AI workflows. Topics covered in this episode: Genkit: An overview of the AI orchestration framework for full-stack applications. Continuous Improvement Loops (Kaizen): Using models like Mistral and Gemini to iteratively refine AI outputs. Vertex AI Model Garden: Accessing serverless models without managing infrastructure. Operational Transparency: Exploring how the Gemini CLI streams thinking tokens to improve user experience. Developer Portfolio Rebuilds: Rody shares his process for rebuilding his website using AI Studio, React, and Astro. Automating Metadata: Embedding Gemma 3N to auto-generate blog descriptions and surface related posts via KNN queries. Astro Galleries: Creating photo galleries with masonry layouts and lightboxes.

Watch on YouTube
Are your users staring at a frozen screen while your massive LLM processes a prompt? In AI UX, a frozen screen means a broken app. Today, we are fixing that by turning your AI's "black box" into a "glass box" using Operational Transparency. In this video, I’ll show you how to eliminate AI wait anxiety by streaming the model's internal "thoughts" directly to your UI in real-time. Drawing on real-world transit psychology from the London Bus system, we dive into the code to show you exactly how to intercept LLM thought signatures using Firebase AI Logic and the Gemini API. Keep your users engaged, build system credibility, and drastically improve your app's user experience.
firebase • Mar 6, 2026

Cure AI Wait Anxiety with the London Bus Hack

Are your users staring at a frozen screen while your massive LLM processes a prompt? In AI UX, a frozen screen means a broken app. Today, we are fixing that by turning your AI's "black box" into a "glass box" using Operational Transparency. In this video, I’ll show you how to eliminate AI wait anxiety by streaming the model's internal "thoughts" directly to your UI in real-time. Drawing on real-world transit psychology from the London Bus system, we dive into the code to show you exactly how to intercept LLM thought signatures using Firebase AI Logic and the Gemini API. Keep your users engaged, build system credibility, and drastically improve your app's user experience.

Watch on YouTube
Leverage Chrome’s on-device Prompt API to deliver a private, infrastructure-free LLM experience. While on-device models like Gemini Nano offer improved privacy by keeping data local, hybrid AI experiences often require a fallback to a cloud model when local resources are unavailable. This post details how to implement essential transparency warnings and give users a choice to proceed, ensuring reliable functionality while maintaining user trust and preserving privacy as your core value proposition.
firebase • Mar 5, 2026

The Privacy Guardrail: How to Implement Warnings for On-Device LLM Failures

Leverage Chrome’s on-device Prompt API to deliver a private, infrastructure-free LLM experience. While on-device models like Gemini Nano offer improved privacy by keeping data local, hybrid AI experiences often require a fallback to a cloud model when local resources are unavailable. This post details how to implement essential transparency warnings and give users a choice to proceed, ensuring reliable functionality while maintaining user trust and preserving privacy as your core value proposition.

Dive in
Ready to move beyond single-model agents? Dive into continuous improvement loops (Kaizen) using Genkit and Vertex AI Model Garden. We show you how to use models from multiple providers as a 'writer' and a 'critic' to build AI agents that critique and refine their own outputs for maximum quality and improved results.
genkit • Mar 4, 2026

Maximize Your Agent's Output: Leveraging Multi-Provider Models in Genkit

Ready to move beyond single-model agents? Dive into continuous improvement loops (Kaizen) using Genkit and Vertex AI Model Garden. We show you how to use models from multiple providers as a 'writer' and a 'critic' to build AI agents that critique and refine their own outputs for maximum quality and improved results.

Dive in