ml-intern vs Vera

ml-intern is an open-source autonomous ML engineering agent from HuggingFace that can read research papers, design experiments, write and run training code, evaluate results, and push trained models to the HuggingFace Hub — all without human handholding. It runs a closed agentic loop for up to 300 iterations, integrating natively with HF Datasets, Inference Endpoints, and documentation. The system includes a doom-loop detector to prevent infinite debugging spirals, session upload to HF for persistent multi-day runs, and supports both zero-shot paper-to-model tasks and structured experiment pipelines. It's specifically designed to run on HuggingFace's own compute infrastructure, which gives it native access to GPU clusters that most comparable agents have to provision externally. The project targets ML researchers and small teams who want to explore a paper's ideas without doing the full implementation grind themselves. The HuggingFace ecosystem integration is the key differentiator — this isn't a generic code agent that happens to write PyTorch; it's purpose-built for the HF workflow, complete with automatic model cards and benchmark uploads.

Vera is a programming language built from the ground up for LLMs to write — not humans. Named after the Latin word for truth, it compiles to WebAssembly and runs in both the CLI and browser. Its most radical design choice: it eliminates variable names entirely, replacing them with typed De Bruijn structural references (like `@Int.0` for the most recent integer binding). Research suggests naming confusion is one of the biggest failure modes in AI-generated code — Vera removes the problem at the language level. Every function in Vera must declare `requires()` preconditions, `ensures()` postconditions, and `effects()` side-effect declarations. The compiler uses Z3 formal verification to check contracts at every call site, meaning the AI can't ship code that violates its own preconditions. Error messages are structured JSON with stable codes — written as instructions for AI systems to parse and fix, not human developers to read. Benchmark results are striking: on VeraBench, Kimi K2.5 achieves 100% correctness writing Vera code, outperforming both Python (86%) and TypeScript (91%) implementations. At v0.0.127 with 810+ commits, 127 releases, 3,638 tests, and a 13-chapter spec, this is a serious project — not a weekend experiment. If AI is going to write most of our code, perhaps the code should be designed for AI to write.