BLOGS

Thoughts, notes & things I've learned

Writing about tech, ideas, and whatever I find worth sharing.

5 posts

Open Source LLMs Aren't Free: The Break-Even Math of Self-Hosting

Open Source LLMs Aren't Free: The Break-Even Math of Self-Hosting

Oct 2026  ·  6 min read

The reasons to self-host an LLM, the break-even formula step by step, and the hidden costs that move it.

LLMSelf-HostingCost AnalysisGPU
Cropping is all you need in OCR, not prompt tuning

Cropping is all you need in OCR, not prompt tuning

Oct 2026  ·  6 min read

A vision model kept misreading handwritten digits. Three prompt rewrites changed nothing. Cutting out each field and enlarging it took critical fields from 58% to 83%.

OCRVision LLMPrompt EngineeringEvaluationComputer Vision
How to read an AI's thoughts before it speaks

How to read an AI's thoughts before it speaks

Jun 2026  ·  4 min read

Anthropic built a tool that translates Claude's internal numbers into readable text. When they tested it on a safety scenario, Claude's own thoughts revealed it knew it was being tested the whole time.

AI SafetyAnthropicClaudeInterpretability
Is NVIDIA NIM's free tier good enough for a real-time voice agent demo?

Is NVIDIA NIM's free tier good enough for a real-time voice agent demo?

Jun 2026  ·  4 min read

NVIDIA NIM gives you free hosted STT, LLM, and TTS under one API key. I plugged it into Pipecat and built a real-time voice agent with VAD, smart turn detection, and idle reminders in a weekend. Here's what the numbers actually look like.

Voice AgentsNVIDIAPipecatReal-timeSTTTTS
How WebRTC Actually Works, All in one post

How WebRTC Actually Works, All in one post

Jun 2026  ·  14 min read

Before any audio or video flows, WebRTC does a lot of work upfront to find the best candidate pair. This post walks you through exactly how that works, step by step.

WebRTCReal-timeNetworkingVoice Agents