Back to Repository

LibriScan

Applied AI
Role: Lead System EngineerYear: 2025

Client project: Harvard Graduate Capstone (Amesbury Library Consortium)

Technical Deep Dive

Led stakeholder-driven system design and shipped a human-in-the-loop digitization platform with Django + Huey orchestration, RBAC, and immutable provenance.

Client Context

Amesbury Library needed an affordable workflow to digitize archaic handwritten manuscripts without sacrificing archival correctness.

Execution

Designed Upload -> Textract -> Review -> Approval state transitions, implemented Django + HTMX review tooling with OpenSeadragon manuscript inspection, and tracked every word edit with reviewer attribution and immutable history.

Outcome

Reduced manual transcription effort by roughly 90% while sustaining around 85% baseline OCR accuracy backed by governed human review and audit-ready provenance.

Reduced manual transcription by ~90% with AWS Textract and achieved ~85% accuracy via human-in-the-loop review.

Core Stack

AWS Textract
Django
Huey
HTMX
OpenSeadragon

Metrics

reduction

90%

accuracy

85%

workflow

Human-in-the-loop