Livio Grid · Open Source

aidc-proposal-parser

The TypeScript library that powers the parsing layer underneath grid.golivio.com/review. Open-sourced because every offtaker, broker, and capital partner deserves a free, accurate parser for AI data center developer proposals.

View on GitHub →npm package Engineering methodology

Install

npm install aidc-proposal-parser

Usage

import { parseProposal } from 'aidc-proposal-parser'

const result = await parseProposal('./developer-proposal.pdf')

console.log(result)
// {
//   capacity: { itLoadMW: 25, totalLoadMW: 31.25, source: 'page-3' },
//   schedule: { rfsMonths: 9, source: 'page-12' },
//   cost:     { totalUsdM: 145, perKwIt: 5800, source: 'page-18' },
//   equipment: [
//     { category: 'UPS', vendor: 'Vertiv', model: 'Liebert APM',
//       leadTimeWeeks: '12', source: 'page-22' },
//     ...
//   ],
//   location: { city: 'Houston', state: 'TX', country: 'US' },
//   tier:     { claim: 'Tier 3', topology: 'N+1' },
//   pue:      { claim: 1.15, mode: 'directional' },
//   completeness: 0.78,
//   missingFields: ['epoLocations', 'fireSuppressionType']
// }

Why a dedicated parser?

AI DC proposals are a specific document genre — different from generic engineering proposals. Equipment names follow vendor catalogs (Vertiv, Schneider, Cummins, GE, Cerebras, Nvidia). Capacity is always quoted in two figures (IT load and total load) that must reconcile via PUE. Schedule is dominated by long-lead procurement, not construction. Site geography determines half the design choices.

A general-purpose PDF parser will get 30% of the structured data right. A purpose-built parser gets 85%+.

What it does NOT do

This is a parser, not an evaluator. It extracts structured claims; it does not judge whether those claims are deliverable.

For a first-principles engineering review of the parsed claims — capacity feasibility, schedule realism, cost benchmarking, NEC/NFPA compliance — use grid.golivio.com/review (free, no signup).

Contributing

The most valuable contributions are new proposal patterns — anonymized real proposals you've seen that the parser handles poorly. Open an issue using the new_proposal_pattern template on GitHub.

License

MIT. See LICENSE on GitHub.