AI-Assisted Data Migration & Governance
A safety-first migration of 6,041 files / 47.4 GiB with no source changes or overwrites, verified by counts, bytes, checksums, and Unicode-aware paths.
- Files migrated
- 6,041
- Total size
- ~47.4 GiB
- Source modifications
- 0
- Remote deletes
- 0
- Remote overwrites
- 0
- Python
- Bash
- rclone
- Google Drive API
- SHA-256/MD5
- Git/GitHub
Problem
A personal digital archive accumulated on one external device needed redundant cloud storage without modifying the original, silently losing files, or uploading material that should remain local.
A bare sync could move the files, but it could not prove that nothing had been lost or mishandled.
Approach
The migration ran as a copy-only, human-in-the-loop workflow. An AI agent assisted with inventory, classification, batching, and verification while a person approved each judgment point.
- Read-only inventory: establish file, byte, extension, and hidden-file baselines.
- Sensitive candidate review: flag credential-shaped names and extensions for human review.
- Explicit exclusions: keep private material on the source device by recorded decision.
- Visible batched transfer: copy in observable increments.
- Full verification: compare counts, bytes, checksums, and NFC/NFD-aware paths.
- Final record: preserve totals, verification results, and intentional exclusions.
Result
- 6,041 files / ~47.4 GiB migrated
- Zero source modifications, remote deletes, or overwrites
- File-count, byte-count, and checksum verification passed
- Sensitive exclusions and final evidence recorded
What this demonstrates
- Designing a workflow that prevents irreversible actions by construction
- Treating verification as a required delivery stage
- Placing human approval at genuine judgment points
- Handling Unicode normalization differences explicitly across filesystems