Applied intelligence,
learned properly.
HumanV is a studio and school for practical large language models — teaching how to build with LLMs in software, and apply them across scientific work. From first prompt to production.
A workspace
and a curriculum.
One place to talk to frontier models — and one curriculum to teach you how to use them well in code and in research. Open the Studio to begin; follow a track to finish.
HumanV Studio
The live workspace — talk to frontier models side-by-side, save sessions, and switch across programming and science presets in one tab.
Open Studio → TrackProgramming with LLMs
From first prompt to production agents: coding assistants, tool use, evals, retrieval, and how to keep LLM code honest in a real codebase.
Start track → TrackScience with LLMs
Reproducible literature review, data analysis, and reasoning workflows for research — with notes on where LLMs help, and where they mislead.
Start track → DevelopersAPI Platform
Programmatic access to the same models in the Studio. Streaming, tool calls, structured outputs, and a free tier large enough to ship a real project.
View docs →Guides & Walkthroughs
Short, opinionated recipes — prompt patterns, eval design, retrieval tactics, and the trade-offs behind each. Every example runs in the Studio.
Browse guides →Community
Weekly study groups, peer project reviews, and a questions channel staffed by the people who write the tracks. Open to builders at any level.
Join community →What we learned
last month.
Prompt patterns for deterministic tool calling
Six repeatable patterns that keep tool-calling agents robust across model upgrades — with eval sets and failure modes for each.
Programming · AgentsEval-driven development for production LLM apps
How to ship LLM features the way you ship tests: red, green, refactor — applied to natural-language behaviour in a real codebase.
Programming · EvalsLLMs in reproducible scientific workflows
Where LLMs accelerate literature review and analysis — and the audit steps that keep them from quietly introducing errors into research.
Science · ReproducibilityTeaching LLMs means teaching where they fail.
A curriculum that only shows what works is dishonest. Every track on HumanV is built on three commitments — to learners, to the people they build for, and to the field.
We show what fails, too.
Every track documents the failure modes a model has — hallucinations, confident-but-wrong code, stale retrieval — and what to do about each.
Every example runs.
No screenshots of prompts. Every snippet in the curriculum is executable in the Studio and versioned against the model it was written for.
Curriculum free to read.
The reading is free. The Studio and API have generous free tiers. We charge for teams and throughput — never for someone trying to learn.