Laying the Tracks for Text Transformation Grounded in Tradition.
Railroad is a structured, open reference system for classical Buddhist texts. Instead of asking AI models to interpret ancient texts from scratch every time, Railroad pre-resolves all the hard interpretive questions — word by word, verse by verse — using the classical commentary tradition. The result is a set of context packages that any AI model can use to produce reliable translations, lessons, or adaptations without redoing the philological work.
LLMs are bad at translating classical Buddhist texts reliably. Not because they lack knowledge, but because these texts are deeply ambiguous — a single Sanskrit compound can have multiple valid readings, and only the commentary tradition tells you which one applies. Feeding raw commentaries into a chat session doesn't solve this: the context is too large, too unstructured, and every translation starts from zero.
Railroad fixes this by doing the interpretive work once and storing it in a structured, reusable format.
For each verse or passage, a context package is compiled containing:
- Source text — original language with transliteration
- Morphological analysis — every word tagged with lemma, part of speech, and compound breakdown
- Syntactic tree — sentence structure following the commentarial reading
- Semantic glosses — the active meaning of each term (and which meanings are excluded)
- Interpretive notes — what each commentary says is happening, with citations
Every field traces back to a specific commentary passage. When commentaries disagree, the disagreement is recorded explicitly — never flattened.
Railroad/
├── 0_Project/ ← project docs and guidelines
├── 1_Human_Sources/ ← ground truth texts (read-only for AI)
├── 2_Authoritative_Context/ ← structured context packages (AI-compiled)
├── 3_Text_Transformations/ ← generated translations, lessons, etc.
└── 4_Wiki/ ← cross-text concept knowledge
Citation chain: 1_Human_Sources → 2_Authoritative_Context → 3_Text_Transformations / 4_Wiki. No link can be skipped. Every AI-generated claim is traceable back to a human source.
| Text | Language | Period | Scope |
|---|---|---|---|
| Bodhicaryavatara (BCA) | Sanskrit | 8th c. | ~900 verses on the bodhisattva path |
| Abhidhamma Pitaka | Pali | Canonical | Seven treatises of systematic analysis |
Two roles are needed:
- Domain Specialists — scholars who evaluate and correct the AI-generated analysis against the commentary tradition. No coding required.
- AI Skills Designers — contributors who build prompts, lint passes, and workflows. No Sanskrit/Pali required.
Start with 0_Project/README.md for full onboarding details, and CLAUDE.md for the schema the AI works from.
- Not an AI model — it's the track, not the engine
- Not a translation archive — outputs are grounded in tradition but are not scholarly editions
- Not finished — coverage is incremental; only
status: completefields are fully reviewed