The Journey
Your path from first commit to capstone
The Journey
A guided, narrative-driven path through your first break in AI.
Audio plays as you scroll through each scene.
AI-generated podcast based on actual cohort material by FireHacker (Amardeep). Learner names are anonymized. Voices and conversation are generated by AI.
Ship something real
The first act of agency
Every journey starts with a commit. You set up a Quarto blog, push it to GitHub, and use an AI IDE to write your first pages. It's not about the blog — it's about proving to yourself that you can ship something real, in the open, with tools that didn't exist two years ago.
Branches, pull requests, merge conflicts — these stop being abstract concepts and become muscle memory. The AI doesn't write your code for you; it writes code with you.
See inside the machine
The demystification moment
You download 3 GB of weights and run Qwen3 0.6B in pure C. No frameworks, no abstractions — just tokens flowing through attention heads, a KV cache growing with every step, and a sampling function picking the next word.
For the first time, an LLM isn't a black box. You can trace every operation from BPE tokenization to chat template parsing to the forward pass itself. Temperature stops being a slider and becomes a probability distribution you can reason about.
Think at production scale
The systems mindset shift
Running one model on your laptop is a start. Serving thousands of requests per second is a different problem entirely. You learn how vLLM and TGI use continuous batching to keep GPUs saturated, how quantization trades precision for speed, and how speculative decoding makes large models respond faster by using a small draft model.
This is where you stop thinking like a user and start thinking like an engineer. The question shifts from "how do I run this?" to "how do I serve this to a million people?"
Train your own
The builder threshold
You cross from consumer to creator. PyTorch tensors, autograd, and training loops become your daily tools. You fine-tune with LoRA, scale with DDP, and climb the parallelism ladder — from a single GPU to tensor, pipeline, and expert parallelism across nodes.
The nanoGPT speedrun becomes your proving ground: train GPT-2 to a target validation loss as fast as possible. You read production code in Megatron and Picotron. The Muon optimizer stops being a leaderboard trick and shows up in production checkpoints.
Ship a product
The product lens
Knowledge without application is trivia. You take everything you've learned — inference, training, systems — and ship an AI-powered product end to end. Problem, solution, users. RAG, agents, tool use. Frontend, backend, deploy, monitor.
This is where the cohort gets real. You're not completing exercises — you're building something people can use.
Prove it
The portfolio signal
A shipped capstone project or a merged open-source PR is the strongest signal on your profile. Not a certificate — a deployed product or a contribution reviewed and accepted by maintainers of an AI project you care about.
You present your work to the cohort. You get peer feedback. And then you have something concrete to point to when someone asks: "What have you built?"
Your first break starts here
This cohort is free, community-driven, and open. Join the Discord, pick up the roadmap, and start building in the open.