Office Hours

Live weekly sessions where you can ask anything, demo your work, and learn alongside the cohort and AI agents.

What happens in office hours

Office hours are live, interactive sessions led by FireHacker. Each session covers a mix of:

• Concept deep dives — transformer architecture, KV cache, training loops, inference engines, and agentic AI patterns explained with diagrams and live code
• Live demos — running Qwen3 in pure C, DDP training, nanoGPT speedruns, MCP tool building, and more
• Learner questions — bring anything you're stuck on. Past topics have included GitHub conflicts, quantization choices, HuggingFace model cards, and RoPE
• Project reviews — share your blog, your model experiments, or your AI product idea for live feedback

Recordings and AI-powered transcripts

Every office hours session is recorded and published with an interactive, searchable transcript. You can click any line in the transcript to jump to that moment in the video. The transcripts are AI-generated and organized by topic — so you can find "KV cache" or "speculative decoding" without scrubbing through an hour of video.

AI agents in the cohort's MCP server can also search office hours content. Ask firstbreakai ask "What was discussed about DDP?" and the AI agent finds the relevant sessions and topics.

Past sessions

Cohort 01 sessions so far have covered transformer architecture, Qwen3 0.6B inference, HuggingFace ecosystem, model formats (GGUF vs SafeTensors), distributed training (DDP), the nanoGPT speedrun, Claude Code skills and sub-agents, and benchmark literacy. Each session page has a full topic index with video timestamps.