Institutional Project Brief · CORPUS-03

Hmar Corpus Archival Project

The Hmar Corpus Archival Project is the primary backend data engine of the Foundation. It operates as the raw ingestion pipeline, dataset schema authority, and educational training ground for text corpora, speech audio archives, and OCR scanned page datasets mirrored on Hugging Face.

Proposed Technical Specifications

SpecificationIntended Details
Archival PortalProject Not Initialized
Staging & PR Workspacegithub.com/hmar-heritage
Master Storage Hubhuggingface.co/hmar-heritage
Contribution ModelAnonymous Simple Upload · Optional Accounts for History Storage

Backend Data Engine & Ingestion Pipeline

Unlike the read-only public storefront of the hmar digital library, the Hmar Corpus Archival Project functions as the backend data management hub. It provides the core data ingestion, storage, and processing pipelines required to handle raw text dumps, high-resolution document scans, and uncurated speech audio recordings.

GitHub Staging Pipeline & Hugging Face Master Storage

The project architecture enforces a clear division between staging workflows and master storage. GitHub is used for all pull requests, code reviews, and in-review staging branches because of its intuitive interface, file diff tools, and issue tracking. Automated GitHub Actions run verification checks and render pending submissions live on the storefront under an in-review status. Once a pull request is merged, automated CI and CD workflows sync and push the verified master release to the Hugging Face Datasets hub for bulk hosting.

Low-Friction Uploads & History Storage Accounts

The portal main page is designed for ultra-simple community participation. Anonymous users can upload scans, audio files, or text transcriptions instantly through a basic web upload interface without creating an account. Optional contributor accounts serve exclusively to store personal submission history and attribution records, containing zero administrative or management features to avoid unnecessary website overhead.

Training Ground for Technical Repository Management

Beyond simple web uploads, the portal serves as an educational training ground for less tech-savvy volunteers. Through step-by-step guides, contributors graduate from basic drag-and-drop web uploads to terminal commands, Git version control, and working directly on the Hugging Face platform. This builds long-term technical capacity within the community.

Data Schema Specifications and Organizational Guides

The project serves as the formal authority for dataset structure. It defines standardized JSON schemas, metadata fields, and scanning resolution benchmarks for text, image, and audio archives. Detailed how-to guides instruct partner organizations, such as churches, schools, and literature societies, on preparing institutional records for archival ingestion.

Planned Technical Deliverables

  • GitHub pull request and staging branch workflow for in-review contribution rendering
  • Automated CI and CD pipelines syncing merged GitHub PRs to Hugging Face master datasets
  • Low-friction anonymous upload interface paired with attribution history accounts
  • Educational onboarding guides transitioning volunteers from simple web uploads to terminal & Git workflows
  • Standardized dataset JSON schemas and digitizing guides for partner institutions