PROBLEM
Real-world OCR fails where reality gets ugly.
Receipts, labels, and scanned documents are messy, damaged, inconsistent, folded, cropped, blurry, and context-dependent.
NW
NULLWORKS
Mason Perry // Operational Systems Builder
OCR / Document Intelligence
Feed messy documents to a system that learns.
PAPERGOBLIN is a live OCR/document-intake system that turns receipts, labels, and scanned chaos into editable correction bubbles, structured operational packets, and telemetry the system can learn from.
OCR recoveryEditable UICorrection telemetryAgent-readable packets
Selected System
Live prototype showing OCR recovery, correction workflows, telemetry, and product-system thinking.
SOLUTION
PAPERGOBLIN converts chaos into packets.
OCR output becomes editable correction bubbles, semantic labels, confidence flags, and structured operational data.
CORE INSIGHT
The human is not fixing OCR.
The human is labeling reality. Every correction becomes training signal for future parsing, routing, and automation.
Workflow Route
Chaos becomes structured signal.
01
scan
02
OCR
03
correct
04
label
05
packet
06
telemetry
07
learn
Built With
Next.js
TypeScript
Tailwind
Tesseract.js
Supabase
Vercel
Why It Matters
I build the operational layer around AI failure.
The model is impressive. The real-world workflow around the model is usually the dumpster fire. PAPERGOBLIN proves the missing layer: correction, validation, confidence, persistence, telemetry, and recovery.
Outcome
Built and deployed during commercial airline travel using laptop + phone + in-flight constraints.
Not a mockup. A working field prototype built under imperfect conditions — because operational systems should survive reality.