Deal Room Reconciler

Local-first Uliya Skill that cross-checks contracts, invoices, receipt photos, and meeting recordings on UGREEN AI NAS to catch discrepancies and organize verified archives privately.

Comment

Inspiration

Freelancers, independent consultants, and small finance or legal teams store confidential contracts, vendor bills, receipt photos, and client call recordings on their NAS. Cross-checking numbers and deadlines across different file formats by hand takes hours, while uploading sensitive financial records to cloud AI services creates major privacy risks.

What it does

Deal Room Reconciler (deal-room-reconciler) is a local-first Uliya Skill for UGREEN AI NAS that audits mixed-media project folders entirely on-device:

  • Inventories & Fingerprints: Lists all files in a target project folder and computes SHA-256 checksums for tamper-evident tracking.
  • Transcribes Meeting Audio: Extracts spoken approvals, price changes, and revised deadlines with timestamps from call recordings.
  • Reads Scanned Receipts: Uses local vision OCR to pull vendor names, dates, taxes, and totals from receipt images.
  • Cross-Reconciles Terms: Compares contracts, invoices, OCR receipts, and call transcripts to flag amount mismatches (AMOUNT_CONFLICT), date conflicts (DATE_CONFLICT), missing supporting records (MISSING_EVIDENCE), and unreadable files (COVERAGE_GAP).
  • Organizes with Approval: Writes a structured reconciliation-report.md and—only after explicit user confirmation—tags and sorts files into verified-archive/ and needs-review/.

How we built it

We designed a five-step sequential orchestration workflow using the Uliya Skill Development Standard (SKILL.md + references/reconciliation-schema.md) that coordinates four native UGOS expert agents (file-agent, knowledge-agent, photo-agent, and document-agent) through self-contained delegate instructions:

  1. ug-filesystem (ls, sha256sum) for read-only discovery and hashing.
  2. ug-knowledge (audio-transcription) for meeting audio extraction.
  3. ug-photo (vision-ocr) for scanned receipt parsing.
  4. ug-document (metadata, extract, reconcile, writing) and ug-knowledge (save-text) for multi-source verification and report drafting.
  5. ug-filesystem (mkdir), ug-filemgr (tag-add, tag-add-file), and file-workflow (batch-move) under strict waiting_approval gates.

Challenges we ran into

Designing strict boundary rules so unreadable or encrypted files never skew financial totals—and ensuring every write, tag, or file move pauses for explicit user approval without losing partial read-only findings—required careful state design across all five Uliya protocol statuses (completed, failed, cancelled, waiting_user, waiting_approval).

Accomplishments that we're proud of

  • Zero Cloud Exposure: Sensitive contracts, receipts, and call recordings never leave the local UGREEN AI NAS.
  • Cross-Modal Verification: Bridges audio transcripts, scanned images, and multi-page documents in one unified audit pipeline.
  • Deterministic Safety: Enforces SHA-256 traceability, explicit coverage-gap reporting, and human-in-the-loop approval before modifying a single file.

What's next for Deal Room Reconciler

Next, we plan to validate end-to-end execution across live UGOS expert agents on the UGREEN AI NAS iDX6011 Pro, benchmark performance on large multi-hundred-file deal folders, and add customizable currency/tax tolerance templates for recurring monthly audits.

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Deal Room Reconciler

Built With

  • ai-agents
  • markdown
  • ocr
  • ugos
  • ugreen-ai-nas
  • uliya
Share this project:

Updates

Submission history