#90DayAIEngineer · your own 92-day calendar
Pick your start day, ship something every week, and finish 92 days later with a portfolio instead of a resolution. Twelve projects, about two hours a day, every video and resource in one place. Your calendar, ship dates and tracker all count from the day you choose.
Before you start
Yes, if you already write basic code in any language: loops, functions, a little Git. New to coding? Keep the plan but slow it down: take Week 0 seriously, add a month of Python, then follow the same order.
"AI engineer" here means someone who builds dependable products on top of existing models: chatbots that stay grounded, agents that finish the job, search that understands meaning. It is not ML research, and you won't be designing new architectures.
The structure
Each month of your plan ends with a checkpoint and each week ends with a project. Twelve weeks, twelve ships.
How LLMs behave and how to talk to them from code.
Retrieval, tools, agents and, above all, evals.
Deploy, monitor, build a capstone and launch it in public.
The only daily rule
One video, one doc page or one chapter. Not five tabs. Take notes by hand.
A tutorial only counts once you've committed code after it.
Every Friday, post what you shipped: a screenshot, a lesson, a failure. Recruiters look for builders, not certificates.
Roughly 2 hours on weekdays and 4 on weekends. Consistency beats intensity.
Week 0 · the 7 days before Day 1 · Setup
The full roadmap
Pick a week to see what to learn, what to ship and the videos to watch. Ticks save in this browser. Rule: no week ends without a public GitHub repo.
Video library
Every YouTube link from the article in one table. Filter by month of your plan or copy a link to share with your batch.
| Week | Video | Why watch |
|---|
More playlists
These free CampusX playlists by Nitish Singh cover the same topics in more depth. Each week's panel above also lists the matching CampusX videos.
Don't break the chain
Each day you finish your hour of learning and your hour of building, tap the box. One row per week of your plan; days after today stay locked until they arrive.
Go deeper
The 12-week plan tells you what to ship each week. These two show the bigger map: one for what the job covers, one for how things work underneath.
A clickable visual map of everything an AI engineer is expected to know, from using pre-trained models and prompting to RAG, agents, embeddings and safety. Each node links to free reading. Use it to check you haven't skipped a topic and to see what comes after these 92 days.
A long-form curriculum maintained by Rohit Ghumare and contributors. You hand-write the tokenizer, attention, backprop and the agent loop before using any framework, then move on to LLM engineering, evals, serving, observability and agents. Runnable code in Python, TypeScript, Rust and Julia.
| Your weeks | What you ship | On roadmap.sh, look at | In From Scratch, go to |
|---|---|---|---|
| W1–W2 | Chatbot, token visualizer | Pre-trained models, how LLMs work, OpenAI/Claude APIs | Setup & tooling, software fundamentals, model architecture (optional depth) |
| W3–W4 | Resume extractor, notes search | Prompt engineering, embeddings, vector databases | LLM engineering: prompts, structured output, embeddings |
| W5–W6 | Chat-with-PDF, research agent | RAG, AI agents, tool use / function calling | Retrieval, agent engineering, tools & protocols (MCP) |
| W7–W8 | Eval harness, LoRA tune | Open-source models, fine-tuning, evaluation | Evaluation & testing, training basics |
| W9–W10 | Deployed app, guardrails post | AI safety & ethics, prompt injection, deployment | Serving & deployment, observability & monitoring |
| W11–W12 | Capstone, launch | Projects section for more build ideas | Pick one deep-dive phase for January |
Build from the bottom up
Most self-taught builders skip layer 06. Demos that work on three examples break on the fourth. A test set, a score and a chart of improvements make you look senior.
Stay on the road
Free resource stack
All free except one book.
January 1, 2027
Your progress lives in this browser
Your batch, ticks and tracker are saved on this device only. Copy your progress code here and paste it on the other device to continue there.