by Carlo Perottino
Things I need. Things I want. Things that interest me.
Eight identity & authorization patterns for AI agents calling tools
Head-to-head notebooks pitting LLMs against the classical methods built for the job
Run the same prompt against flagship models side-by-side and see exactly where they differ
Continuous internet speed monitoring with anomaly detection and automated provider accountability
A framework for running socialization experiments between LLMs — games, debates, alignment tests
Elsewhere on GitHub
A few things I’ve made outside Forge—useful, interesting, and still worth sharing.
AI-powered contact management and personalized outreach for your professional network.
A practical AWS workflow for training a personal likeness LoRA from a folder of photos.
A life-event-driven financial simulation tool for long-range scenarios and projections.
A weekend-sized, hands-on path to building and training a complete GPT from raw PyTorch operations.
AI-powered competitive intelligence engine for generating personalized executive outreach messages in B2B sales.
A gallery and workbench that runs one prompt across flagship models, aligns their outputs, and surfaces meaningful differences.
An interactive visualization of the full LLM lifecycle, replaying real training runs from initialization through pre-training, SFT, reinforcement learning, and inference.
A self-hosted toolkit that deploys DNS, ICMP, and HTTPS tunnels on EC2 for resilient connectivity in restricted networks.
Ordered, runnable exercises for studying the OpenAI API stack, from Responses and built-in tools to retrieval, agents, and production patterns.
A local benchmark harness for measuring generated SQL's end-to-end runtime across query-complexity levels and context strategies, using OpenAI models for generation.