Example projects built with Astro -- a planning-first platform for builders and scientists.
Each example includes the planning DAG (exported as .astro.json), a visual DAG diagram, detailed task descriptions with prompts, and the final deliverables produced by Astro's AI-assisted execution pipeline.
Transform research papers into multi-format presentation materials -- slide decks (light & dark themes), long-form blog posts, and conference posters. Astro decomposes the paper into key concepts, generates visuals, and produces publication-ready outputs in parallel.
| # | Paper | Deliverables | DAG |
|---|---|---|---|
| 1 | Physics of Language Models: Part 1 -- Allen-Zhu & Li, ICML 2024 | 40-slide deck (light+dark PDF), blog post, A0 poster | 12 tasks, 15 edges |
| 2 | Physics of Language Models: Parts 4.1 & 4.2 -- Canon Layers & Scaling | 33-slide deck (light+dark PDF), blog post, A0 poster | 10 tasks, 18 edges |
Deep-dive competitive analysis reports for public companies -- evaluating economic moats, competitive advantages, financial metrics (ROIC vs WACC), and scenario modeling. Produces a full investment-grade report with interactive SVG charts.
| # | Company | Deliverables | DAG |
|---|---|---|---|
| 1 | CRWV -- CuriosityStream Moat & Fragility | HTML+PDF report, CA radar, moat heatmap, ROIC vs WACC, scenario chart (5 SVGs) | 13 tasks, 18 edges |
Automated peer-review-style analysis of research papers -- structured evaluation with scoring rubrics, methodology assessment, and reproducibility checks. Outputs a LaTeX-formatted review report with a visual scoring dashboard.
| # | Paper | Deliverables | DAG |
|---|---|---|---|
| 1 | Geometry-Based Activation for Sparse Autoencoders | LaTeX review report with scoring dashboard + PDF | 11 tasks, 16 edges |
Automated daily research digest pipeline -- fetches the day's papers from HuggingFace, ranks by community upvotes, generates a digest of all papers, then produces in-depth analyses with architecture diagrams for the top-ranked papers. Tasks run in parallel after data retrieval.
| # | Date | Papers | Deliverables | DAG |
|---|---|---|---|---|
| 1 | 2026-03-11 | 30 total, 10 in-depth | HTML+MD+PDF digest, compiled report, 10 per-paper analyses, 10 SVG architecture figures | 7 tasks, 8 edges |
academic-presentations/
1-physics-llm-one/
plan/ # DAG visualization + detailed task docs
dag.svg # Visual DAG diagram
plan-detail.md # Full task descriptions and prompts
plan.astro.json # Raw Astro export
deliverables/ # Final outputs (self-contained)
slides/ # 40 HTML slides + combined PDFs
blog/ # Markdown post + figures/
poster/ # HTML + A0 PDF + figures/
moat-analysis/
1-crwv-moat-fragility/
plan/ # DAG visualization + detailed task docs
deliverables/ # HTML + PDF report with SVG charts
assets/ # CA radar, moat heatmap, ROIC, revenue, scenario SVGs
paper-reviews/
1-gba-sparse-autoencoder/
plan/ # DAG visualization + detailed task docs
deliverables/ # Review report (LaTeX + PDF)
daily-paper-analysis/
1-hf-daily-papers-2026-03-11/
plan/ # DAG visualization + detailed task docs
deliverables/ # Digest + compiled report + per-paper analyses
papers/ # 10 individual HTML+MD analyses
figures/ # 10 SVG architecture diagrams
data/ # JSON metadata + ranked paper lists
utils/ # Tools for generating visualizations from .astro.json
render-plan.js # DAG SVG + plan documentation generator
Generate plan visualizations from any .astro.json export:
node utils/render-plan.js path/to/plan.astro.json path/to/output-dir/Produces:
dag.svg-- Visual DAG diagram (consistent style across all projects)plan-detail.md-- Per-task descriptions, dependencies, and promptsplan-summary.md-- Compact table for README embedding
Astro provides an AI-assisted planning layer that sits above execution tools (coding agents, analysis tools) to help users decompose complex goals, manage dependencies across environments, and focus human attention on strategic decisions.