Enter a math modeling competition problem and get a full drafted paper with figures and code automatically.
Use the human review checkpoint to approve or reject the AI-generated draft before final PDF compilation.
Run benchmarks to evaluate pipeline performance and view token usage and timing reports.
Index reference material so the pipeline can retrieve relevant context while building models and writing sections.
| 123-qw-as/beacon | frenzymath/danus | agricidaniel/claude-shorts | |
|---|---|---|---|
| Stars | 92 | 92 | 93 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | hard | moderate |
| Complexity | 4/5 | 5/5 | 3/5 |
| Audience | general | researcher | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Python 3.11+, Node.js 18+, and a user-configured compatible large language model endpoint.
Beacon is a tool built for students who enter math modeling competitions, where teams receive a complex problem and must produce a full written paper within a tight time limit. Instead of leaving students to manage every step by hand, Beacon takes a competition problem as input and walks it through a multi-step pipeline: it analyzes the problem, builds mathematical models, writes and runs code, generates figures, drafts the paper section by section, assembles tables, scores the work against contest rubrics, and compiles the final document to PDF. A human review checkpoint lets a person approve the draft before final compilation, or reject it so the run ends safely without producing a finalized paper. The pipeline is built on a framework called LangGraph, which coordinates fourteen visible stages. Each stage has a specific role. The Analyst breaks the problem into sub-questions, variables, goals, and constraints, then a Blueprint Critic reviews that breakdown. The Modeler constructs models in stages from basic to final, and a Model Critic checks the assumptions and derivations. The Coder generates runnable experiment code, and a consistency check verifies that the code, model, and output metrics align. Later stages handle sensitivity analysis, figure generation with quality review, section-by-section writing, table assembly, rubric-based evaluation, and human review before LaTeX compilation. Beacon offers two interfaces. A browser-based Web UI lets users paste a problem or upload files in formats like PDF, Word, Markdown, or Excel, then watch the pipeline progress in real time with logs and artifact previews. A command-line tool provides the same functionality for terminal users, with commands to start a run, resume after review, recover from a crash, view a report of token usage and timing, index reference material for retrieval, or run a benchmark suite. The project is written in Python for the backend, with a JavaScript frontend. It requires Python 3.11 or higher and Node.js 18 or higher, and it connects to a compatible large language model endpoint that the user configures. The README includes installation steps, configuration notes, and a project structure overview. The full README is longer than what was shown.
Beacon is an AI tool that takes a math modeling competition problem and automatically writes a full paper, it builds models, runs code, makes figures, drafts sections, scores against rubrics, and compiles to PDF with a human review step.
Mainly Python. The stack also includes Python, JavaScript, LangGraph.
The license for this repository is not specified in the available information.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly general.
This repo across BitVibe Labs
double-check against the repo, no cap.