Chatbots can answer. Agents can work. But humans can desire. The desire is to be more curious, and curiosity is what creates this loop.
Imagine → Discover → Learn → Create → Iterate.
Engineermaxxing is a self-evolving learning and research platform that helps you ride this curiosity loop.
01Leveling the field
“What I cannot create, I do not understand and what I do not understand, I cannot teach.”
After Feynman. The rule the site is built on.
AI lets learned experts explore and expand their research, but that alone doesn't level the field. Access to a frontier model doesn't make someone a learned expert. An ecosystem built around those models, fueling human curiosity, can.
Engineermaxxing is an attempt at that. We start with engineering, but the tools retrofit to any domain: systems for multimodal learning, researching, experimenting, and building, all in one platform.
And those systems need to be agentic in the product, agentic in the making.
02Why now
Everyone has an answer machine now. Nobody has a mentor that adapts to you and teaches from first principles.
A tutored student beats a classroom by two grade levels. Bloom measured it in 1984, and it never scaled. The AI that could finally scale it mostly does the opposite: it hands over the answer and takes the understanding with it.
So we build the mentor's job instead. Meet you the way you learn, with eight ways into every lesson. Make you prove it, by building, teaching back, and recalling days later. Keep the syllabus at the frontier, so research lands in the morning and becomes lessons the same day.
Agents are what make that buildable. They can hold a mentor's patience for one learner, and they can build the school itself: writing lessons, reviewing them, shipping them overnight. One-to-one mastery stops being a privilege and becomes infrastructure.
03What's running live
Not a prototype. This site, in daily use and daily development.
Interactive lessons across AI and hardcore engineering: signal processing, robotics, estimation. The field drawn as terrain; Directions plots prerequisite-correct routes.
04Eight ways into a lesson 7 live 1 deploying
People don’t learn one way. Each way has a film. They play as you scroll; tap for sound.
Lessons that read like a story
Deep-dive lessons that build every idea from zero: concrete before abstract, derivations you can follow, no “it is well known that.”
Open a lesson →A lecture in your ears
Every lesson read aloud the way a patient teacher would explain it: clear, unhurried, and easy to follow with your eyes closed.
Press play →Problems that push back
Interactive workbooks across derive, trace, build, design and debug. You play with each concept until it clicks.
Open a workbook →Think on the page
A notebook that travels with you across the site. Capture, annotate and connect ideas as you go.
Open the notebook →The whiteboard test
Teach Mode hands you a marker and a camera. Explain it simply, or discover the gap. The Feynman test, recorded.
Grab the marker →Teach it back, out loud
The teach-it-back studio listens as you explain a concept and coaches your delivery. You don’t really know it until you can present it.
Start a take →Learn shoulder to shoulder
A live shared whiteboard with voice and screen-share. Work a problem through with someone else in real time.
Enter the playground →Curate a journey
Stitch lessons into a path and share it with a single link. Your sequence becomes someone else’s on-ramp.
Build a path →The extras baked in
Soundscapes, a focus timer, themes and typefaces: the small touches that turn studying into a place you want to be.
Watch the film →05The agentic layer in build
Agents on both sides of the glass: a mentor for the learner, a maker for the platform.
The mentor: one ask becomes a planned route: intent parsed, graph walked, gates placed, reasoning visible. Not a chatbot. The maker's story is below, in five films.
Describe it and the agent builds it
Point it at a link, paper, or topic and a drafted lesson comes back. Pipeline built; surface designed.
Watch the film →What’s worth your time, and why
The mydolven layer reads 33+ sources and publishes The Daily Delve every morning, with a narrated brief.
Open today’s edition →Learn it, then run it
Distill a model, train a toy, reproduce a paper. Sandboxed and cost-capped. An in-browser pilot is live.
Watch the film →Every artifact moves through your review
Draft → review → publish. Nothing auto-ships.
Watch the film →The same loop, pointed at hardware
Builds that leave the screen. The perception track is live today.
Watch the film →06The method live
A new shape of company: we decide, agent fleets do the work.
- 25 codified agent pipelines: lesson, paper, track, audio, video, deck. New content means invoking a recipe, not starting over.
- Fleets, not features: rollouts and reviews fan out to parallel agents. One review pass ran 54 agents and fixed 46 findings before ship.
- Autonomous by default: the daily edition researches, writes, narrates, and publishes itself. Overnight runs ship lessons to production.
- Checked, always: agents adversarially test depth, reproduce numbers against scipy, and screenshot every surface before anything commits.
A virtual lab, driven by humans.
07Tell us what you think
If something taught you well, or wasted your time, or should exist and doesn’t, we want to hear it. Or if you just want to say hello.