Fieldseal

Help get the job done correctly. Seal documentation for every job. Private until shared

Comment

Inspiration

Field technicians often complete good work but document it through disconnected forms, photos, messages, and files. This makes later review, handover, and verification unnecessarily difficult.

We wanted to build a worker-first system where documentation is created as part of the job—not reconstructed afterward. Reports should remain private while still giving owners, contractors, and authorities a way to verify that an issued package has not changed.

What it does

FieldSeal guides field work through five connected stages:

  1. Define the assignment and responsibility.
  2. Plan the work with deterministic source considerations and optional local AI assistance.
  3. Review relevant requirements before execution.
  4. Record completed work, evidence, measurements, deviations, and final inspection.
  5. Issue a private report, anchor its commitment on Midnight, and grant controlled access.

The application supports multiple workers while preserving one clearly identified professionally responsible person. Evidence and important events receive automatic timestamps.

Reports remain private in FieldSeal. Only a salted commitment and minimum receipt state are sent to Midnight. Recipients can receive purpose-bound, section-limited, time-limited, and revocable access without becoming members of the issuing organization.

FieldSeal does not certify work, determine competence, or claim regulatory compliance. Qualified people and the relevant authority having jurisdiction remain responsible for those decisions.

How we built it

FieldSeal is a Python and Flask application with a restrained JavaScript interface and SQLite-backed workflow state.

The documentation workflow separates deterministic rules from AI suggestions. A locally operated Gemma model receives a deliberately restricted technical planning payload. Customer names, addresses, organizations, contact information, evidence files, recipients, and report manifests are excluded. Suggestions remain editable and cannot approve or authorize work.

The US demonstration uses a versioned planning index referencing OSHA, Florida Building Commission resources, NFPA 70, applicable ANSI/TIA standards, manufacturer instructions, and local AHJ verification. It is explicitly a planning aid rather than a compliance engine.

When a report is issued, FieldSeal:

  • Freezes the accepted plan and final-inspection record.
  • Encrypts and stores the private package.
  • Creates a salted document commitment.
  • Registers the commitment through a Compact smart contract on Midnight Preprod.
  • Records the confirming transaction and block.
  • Allows the unchanged package to be verified later.
  • Keeps report contents and personal data off-chain.

The repository also includes end-to-end workflow tests, authorization checks, upload validation, dependency audits, backup verification, and Midnight worker integration.

Challenges we ran into

The main challenge was preserving useful proof without publishing sensitive field documentation.

We had to clearly separate three different concepts:

  • Professional approval of the work.
  • Application-level access to private report sections.
  • Blockchain verification of an unchanged commitment.

Midnight can verify the registered commitment, but it does not prove workmanship, measurement accuracy, identity, competence, inspection, or compliance. Keeping that distinction clear in both the interface and data model was essential.

Another challenge was adapting a Norwegian product foundation into a reviewable US hackathon demonstration without mixing the two delivery lines. We preserved Git history, documented the shared lineage, introduced a separate US source pack, and excluded production databases, evidence, credentials, wallet seeds, and private state.

Accomplishments that we're proud of

  • A complete worker journey from assignment to archived report.
  • Private issued packages with independently verifiable commitments.
  • Purpose-bound and revocable report sharing.
  • Section-level disclosure enforced without altering the anchored package.
  • Automatic timestamps and access history.
  • Local AI assistance with a strict personal-data boundary.
  • Profession-specific measurements and final-inspection records.
  • A confirmed Midnight Preprod anchoring flow.
  • Security and workflow tests covering roles, invitations, revocation, tampering, uploads, and AI boundaries.
  • A transparent provenance document describing the hackathon contribution and its existing foundation.

What we learned

Blockchain is most useful here as a narrow trust layer, not as a document store.

Keeping the complete report private while anchoring a salted commitment provides a clearer privacy boundary than publishing documents, measurements, or personal information. It also allows later recipients to verify integrity without giving them access to the issuing organization.

We also learned that AI works best as an explicit documentation assistant. Suggestions must remain separate from deterministic rules, editable by the worker, and unable to claim approval, safety, competence, or compliance.

Finally, the most valuable workflow is one a trusted technician can complete without unnecessary administration, while students and apprentices remain subject to appropriate supervision and review.

What's next for FieldSeal

Our next steps include:

  • Independent security and privacy review.
  • Mobile testing with technicians in real field conditions.
  • Reviewed jurisdiction-specific source packs.
  • Stronger public-key document signatures.
  • Production hardening of the Midnight service wallet and anchoring worker.
  • Organization subscription and platform-capacity controls.
  • Richer owner and authority handover experiences.
  • Controlled agent access for property transactions and due diligence.
  • Possible future integration with agent-payment infrastructure such as Masumi.
  • Deployment of a public synthetic demonstration containing no real customer data.

Built With

Share this project:

Updates