Ownership Lens
An investigation tool for offshore shell company networks built on a graph database. It analyzes networks of entity relationships through multi-hop graph traversal and graph algorithms to identify potential risk signals.
Timeline (last 90 days)
Ownership Lens is a hackathon project entered in the Graph Hacks hackathon co-hosted by WeMakeDevs and FalkorDB, used to investigate offshore shell company networks.
In the case, the target company GLAISDALE LIMITED showed no direct connection to the watchlist entity DOM Limited at investigation depth 1, and the system returned clear; when the depth was extended to 4 hops, a hidden chain through Rachel Erickson, NEWCOMBE LIMITED, and ANDREW MORAY STUART was found, and the system's verdict was upgraded to escalate.
Ownership Lens uses a real subset of the ICIJ Offshore Leaks database, covering the Samoa and Seychelles jurisdictions in the Panama Papers, and the dataset contains 51,940 nodes and 58,891 relationships.
Ownership Lens's graph model defines relationship types such as REGISTERED_ADDRESS, OFFICER_OF, and INTERMEDIARY_OF.
Ownership Lens's LLM never writes raw Cypher; instead, it chooses from six typed, parameterized, read-only MCP tools: resolve_entity, ownership_chain, shared_links, path_between, cluster_view, and recall_cases.
Ownership Lens's compliance verdicts come entirely from the deterministic Python rule layer in core/rules.py, and the application layer assigns one of four verdicts: clear, review, escalate, or insufficient_evidence.
Ownership Lens's rule signals include officer degree ≥ 5, being in the same community as a watchlist entity, a watchlist path cost ≤ 8.0, and structural anomalies identified by graph algorithms, with thresholds calibrated against empirical percentiles of the dataset.
Ownership Lens's core/verifier.py checks every named entity, executive, and address in the LLM-generated memo one by one against the raw tool output of the current investigation; if any unverified name is found, the system discards the memo and instead generates a deterministic template directly from the evidence ledger.
Ownership Lens deployed the nvidia/llama-3.3-nemotron-super-49b-v1.5 model via a hosted NVIDIA endpoint, benchmarked multiple models across 10 standardized tool-calling scenarios, and passed 15/15 verdict fidelity checks across three multi-turn investigation runs.
The Ownership Lens project ran degree-matched control tests on 30 randomly sampled shell companies, none of which were connected to DOM Limited.
3 more timeline events
All Facts (13)
The Ownership Lens team emphasizes that any real deployment requires human oversight, independent verification, and appropriate governance before it can be used to support consequential compliance decisions.
50%UnverifiedOwnership Lens is a hackathon project entered in the Graph Hacks hackathon co-hosted by WeMakeDevs and FalkorDB, used to investigate offshore shell company networks.
50%UnverifiedOwnership Lens uses a real subset of the ICIJ Offshore Leaks database, covering the Samoa and Seychelles jurisdictions in the Panama Papers, and the dataset contains 51,940 nodes and 58,891 relationships.
50%UnverifiedOwnership Lens's graph model defines relationship types such as REGISTERED_ADDRESS, OFFICER_OF, and INTERMEDIARY_OF.
50%UnverifiedOwnership Lens's LLM never writes raw Cypher; instead, it chooses from six typed, parameterized, read-only MCP tools: resolve_entity, ownership_chain, shared_links, path_between, cluster_view, and recall_cases.
50%UnverifiedOwnership Lens's compliance verdicts come entirely from the deterministic Python rule layer in core/rules.py, and the application layer assigns one of four verdicts: clear, review, escalate, or insufficient_evidence.
50%UnverifiedOwnership Lens's rule signals include officer degree ≥ 5, being in the same community as a watchlist entity, a watchlist path cost ≤ 8.0, and structural anomalies identified by graph algorithms, with thresholds calibrated against empirical percentiles of the dataset.
50%UnverifiedOwnership Lens's core/verifier.py checks every named entity, executive, and address in the LLM-generated memo one by one against the raw tool output of the current investigation; if any unverified name is found, the system discards the memo and instead generates a deterministic template directly from the evidence ledger.
50%UnverifiedOwnership Lens deployed the nvidia/llama-3.3-nemotron-super-49b-v1.5 model via a hosted NVIDIA endpoint, benchmarked multiple models across 10 standardized tool-calling scenarios, and passed 15/15 verdict fidelity checks across three multi-turn investigation runs.
50%UnverifiedThe Ownership Lens project ran degree-matched control tests on 30 randomly sampled shell companies, none of which were connected to DOM Limited.
50%UnverifiedOwnership Lens can launch the full stack via Docker Compose (including FalkorDB, the browser interface, and the FastAPI application), serving an interactive Cytoscape graph canvas at localhost:8000.
50%UnverifiedA clear verdict from Ownership Lens only means that no qualifying indicators were found under the configured rules and search scope; it does not guarantee that the entity is risk-free.
50%UnverifiedIn the case, the target company GLAISDALE LIMITED showed no direct connection to the watchlist entity DOM Limited at investigation depth 1, and the system returned clear; when the depth was extended to 4 hops, a hidden chain through Rachel Erickson, NEWCOMBE LIMITED, and ANDREW MORAY STUART was found, and the system's verdict was upgraded to escalate.
50%