Tag
87 posts
I built a 49-prompt test suite to evaluate Gemini 3.1 Flash Lite's audio understanding capabilities across 13 categories — from accent detection to deception analysis. Here's what worked, what didn't, and why it matters.
A Python tool that connects to remote systems via SSH, probes hardware specs, and generates detailed reports with AI workload upgrade recommendations.
A comprehensive collection of Hebrew language AI models on Hugging Face, covering LLMs, speech recognition, sentiment analysis, and more.
A detailed look at my constantly evolving AI stack, from LLM APIs and frontends to vector storage, orchestration, and Docker deployment.
A structured context data repository of career information designed to power AI assistants for cover letters, resume tailoring, and job search support.
A demonstration of how to structure a public context data repository for seeding LLM-based assistants with personal information.
Using AI to brainstorm and seed personal context data stores that make LLM interactions exponentially more useful and personalised.
An architecture proposal for generating personalised AI context through deliberate interview-driven workflows rather than passive data collection.
A workflow for building personal context data stores using specialized AI agents that interview you and feed structured data into RAG pipelines.
A massive collection of prompts for rewriting any text in Shakespearean English, organized by format and purpose.
A curated collection of system prompts that transform raw speech-to-text output into polished formats like emails, blog posts, and business documents.
A comprehensive library of system prompts for transforming dictated or raw text into structured formats like emails, docs, and reports.
A repository of general-purpose system prompts for configuring LLMs as useful everyday assistants, with personality and context modules.
A Streamlit app for building custom LLM system prompts by mixing and matching configurable identity, style, and output parameters.
A personalized desktop dashboard combining Hebrew calendar data, zmanim, weather, email, and Google services for daily productivity.
Testing whether Gemini 2.5's 65K token output limit can produce a full book from one prompt. Spoiler: Anthropic did it better.
An experiment generating a 100+ page fictitious travel memoir with Sonnet 3.7 from a single prompt, featuring a magical talking sloth.
Test prompts and notes exploring cultural bias, censorship, and worldview differences across Western and Eastern large language models.
A collection of system prompts for intentionally weird and funny AI characters, designed for speech-to-speech platforms.
A fun experiment setting up two AI agents with conflicting secret missions and watching them try to interrogate each other.
Planning notes for a multi-subagent system that turns AI code generation into structured learning opportunities, not just output.
Exploring the idea of using AI code generation tools not just to write code, but as structured tutors that teach you what they did and why.
An AI-powered tool built with Google AI Studio to help people explore and plan career pivots using Gemini's capabilities.
AI platforms make it hard to filter models by multimodal capabilities. I built an open-source taxonomy that maps which inputs produce which outputs.
I built a body language analysis app using Google AI Studio's vibe coding interface and Gemini's multimodal vision. Upload a photo, get expert-level analysis.
I built an AI agent that goes undercover to test other LLMs - probing for biases, guardrails, knowledge cutoffs, and behavioral patterns.
Claude Code isn't just for writing software. I've been collecting projects that use it for research, writing, budgeting, therapy tracking, and more.
I tested 12 text-to-image models on their ability to render Hebrew. Only 2 out of 12 got both test words right. Here are the results.
Traditional resumes are built for human eyes. I created an open JSON schema that gives AI recruiters and screening agents structured, queryable candidate data.
A Gradio web app that uses AI to recommend existing open source licenses or generate custom ones based on your requirements described in plain English.
An AI-powered React app that analyzes how different countries approach policy challenges, with interactive clustering visualizations powered by Gemini.
An automated pipeline that converts raw voice recordings into polished blog posts using audio preprocessing, Gemini transcription, and AI-powered formatting.
A voice analysis application built with Google AI Studio and the Gemini API, exploring multimodal AI capabilities for audio processing.
A curated list of command-line AI coding tools maintained by the model vendors themselves, from Claude Code to Gemini CLI.
A small curated collection of MCP servers for accessing Jewish texts, libraries, and calendar data through AI agents.
Notes and resources on the pain points of MCP adoption and why consolidation through gateways and connectors matters.
Planning notes and sketches for an MCP gateway architecture that aggregates servers into LAN and WAN gateways.
A Claude Code plugin that systematically prepares human-built codebases for agentic AI development workflows.
A tool for maintaining personalized writing rules and vocabulary preferences, consolidated into a Claude Code slash command.
An experimental AI agent that discovers, categorizes, and visualizes technology ecosystems from a single keyword search.
A curated directory of Model Context Protocol servers providing access to Israeli data sources, government APIs, and services.
A curated index of 100+ voice technology tools accessible to Linux desktop users, from real-time dictation to dev frameworks.
A curated resource list of multimodal AI models with native audio support — models that process audio tokens, not just transcribe.
Comparing 8 STT models on a 27-minute podcast. Local Whisper wins on word accuracy, but cloud APIs dominate punctuation.
A short curated list of the best Whisper fine-tuning resources: tutorials, notebooks, and managed compute examples.
Evaluating whether fine-tuning Whisper improves transcription accuracy. Spoiler: it depends on model size and use case.
A script for fine-tuning OpenAI's Whisper speech recognition models using Modal's serverless GPU infrastructure.
A voice-controlled Linux virtual keyboard using Deepgram's Flux turn-taking STT API, built in Rust.
A GUI tool for collecting audio training data for ASR fine-tuning, with LLM-generated prompts and Hugging Face integration.
An MCP server for audio transcription using multimodal LLMs like Gemini, GPT-4o Audio, and Voxtral — not traditional ASR.
An MCP server that brings Gemini-powered audio transcription directly into Claude Code and Claude Desktop.
A desktop transcription app that sends audio directly to multimodal AI models for single-pass transcription and formatting.
A local voice typing app for Linux/Wayland using NVIDIA's Parakeet model. No cloud, no GPU, built-in punctuation.
A snapshot comparing Hebrew TTS quality across six providers, including voice cloning experiments via Replicate.
A concept for capturing end-of-day work progress via voice memos, processing them with Gemini AI, and delivering morning briefings.
An experiment using AI agents to simulate geopolitical dialogue between state actors, non-state actors, and civil society in the Middle East.
A multi-agent system template for conducting comprehensive software and hardware technology evaluations using Claude Code.
An experimental Model United Nations simulation where AI agents embody country positions, vote on resolutions, and analyze bilateral impacts.
An experimental multi-agent system that simulates expert panel discussions, analyzing complex topics through multiple analytical lenses.
An experiment in perspective synthesis — using AI agents to simulate a conference where diverse personas deliver speeches on AI's impact.
An MCP server that lets multiple Claude Code sessions communicate via encrypted peer-to-peer messaging on localhost or LAN.
A framework-agnostic data model and interactive visualization mapping every layer of a production agentic AI system.
A collection of 93+ slash commands for Claude Code CLI covering system admin, optimization, and configuration on Linux desktops.
Another server management template for Claude Code, focused on general Linux server administration with 38 commands and 10 agents.
A template repo with 38 slash commands and 10 agents for managing Proxmox and home servers through Claude Code CLI.
A template repository with 20+ slash commands and 5 specialized agents for managing Home Assistant OS through Claude Code.
A template repo that provides a structured workflow for evaluating and shortlisting GitHub repositories using Claude Code.
A deterministic slash command for Claude Code that transforms messy file structures into well-organized, maintainable systems.
A pattern for intelligently assembling multi-agent crews in Claude Code while staying within context window limits.
An AI-assisted framework for evaluating business ideas using the ICEC methodology — Impact, Confidence, Ease, and Contextual Suitability.
A Claude Code framework that analyzes decisions through seven frameworks including SWOT, cost-benefit, and regret minimization.
A structured model for conducting deep research with Claude Code, featuring pipeline configs, context management, and slash commands.
A template for using Claude Code and git repos to track therapy sessions, mental health goals, and personal growth.
A repo template that transforms free-form project descriptions into organized specifications Claude Code can reliably parse.
A Claude Code workspace template for workflow planning and time management with specialized agents for scheduling, goal tracking, and task triage.
A Claude Code workspace template for household budget management with specialized agents for expense analysis, goal tracking, and report generation.
A real-world example of using a Claude Code research workspace template for ADHD drug policy research and advocacy.
A Claude Code template that turns a Git repository into a structured policy research environment with specialized agents and workflows.
A structured workspace template for writing projects with Claude Code, built around preserving authorial voice and systematic version control.
A version-controlled workspace template for managing Android devices through ADB with Claude Code CLI assistance.
A proof-of-concept multi-agent system that gathers real-time news intelligence and generates scenario-based geopolitical forecasts from six analytical lenses.
A Docker microservices stack for monitoring Pikud HaOref red alerts with real-time maps, push notifications, AI sitreps, and Home Assistant integration.
Charts and analysis from the Promise Denied OSINT project tracking four rounds of Iranian missile and drone strikes against Israel and coalition targets since April 2024.
Introducing the Claude Code Repos Index — a curated directory of 100+ repositories exploring Claude Code as a multi-purpose agent workspace.
An overview of Promise Denied, an open-source intelligence project documenting Iran's ballistic missile, cruise missile, and drone attacks against Israel.
How I fine-tuned OpenAI's Whisper to handle English-Hebrew code-switching, the natural language mixing that happens when immigrants blend two languages mid-sentence.
A curated index of 75+ Claude Code projects and agent workspace templates spanning sysadmin, legal, health, finance, and more.
