Features Cross-portal projects
Pro tier and above

Cross-portal project grouping

Entity resolution collapses 8,000 annual tenders across 17 portals into approximately 1,500 coherent procurement projects. Instead of tracking duplicate notices, re-advertised lots, and sub-packages separately, follow the project — from planning through PIN to design brief to construction tender to subcontracts.

Start Pro trial Book a demo

The deduplication problem

Pacific-aid infrastructure procurement is fragmented by design. A single power grid upgrade project may produce: a preliminary engineering notice on ADB CMS, a procurement plan update on ADB Documents, an invitation for expressions of interest on AusTender, three lots on Tenderlink (PNG Power, Solomon Power, EFL), a request for qualifications on ADB eTendering, and a subcontract notice on two Tenderlink utility portals. Six portals, eight records, one project.

Without project grouping, capture teams track 8 separate records, receive 8 separate alerts, and miss the connections between them. With cross-portal projects, all 8 records collapse into a single project view — with a timeline showing every stage as it posted, and a unified set of tender documents.

How entity resolution works

  1. 01

    Scope-text embedding similarity

    Tender descriptions are embedded and compared across portals. Records with cosine similarity above a threshold are candidates for merging — weighted against buyer name, country, sector, and indicative budget band.

  2. 02

    Structured metadata matching

    Project numbers (ADB project IDs, World Bank P-numbers, DFAT program IDs) are extracted from tender texts and documents. Exact matches are merged automatically with high confidence.

  3. 03

    Buyer and donor chain graph

    The system maintains a graph of buyer organisations and their donor relationships. A PNG Power tender and an AusTender entry both referencing AIFFP financing are structurally related — this narrows the similarity search space.

  4. 04

    Curator queue for ambiguous cases

    Cases below the confidence threshold are queued for human review. A curator sees the two candidates side-by-side with the similarity signals highlighted, and makes a merge / split / flag decision. Curator decisions feed the model as training signal.

What a project view shows

Project timeline — every linked notice in chronological order (PIN → design → tender → subcontract → award)
All linked tender documents — searchable as a single corpus
Donor and financing source — aggregated across all linked records
Total indicative project value — sum of disclosed lot values + estimate where undisclosed
Confidence of merge — "auto-merged (high)" or "curator-reviewed"
Subscribe to this project — alert when any linked notice posts or amends

Related features

AI Capture Brief → Structured briefs with citation provenance Pre-tender pipeline → 6–18 months before the notice drops