AI for internal reviews.
Find connections across emails and documents. Examine answers alongside the sources they rely on.
Request a demoNEXUS helps teams focus on relevant information within large document sets. The language model runs in your organisation’s environment; final judgement remains with your experts.
On the customer's machine
Open-weights models and local inference. Nothing about the case leaves; no telemetry and no training on client data.
Source-grounded answers
Answers include source-document citations. Refusal when evidence is absent is evaluated through case-specific tests; investigators verify the findings.
Investigation-first
Subjects, counterparties, period and allegation: a versioned case briefing drives everything from search to report.
- 1
Preserve
Catalogue originals, record SHA-256 hashes and reconstruct email threads.
- 2
Investigate
Language-aware full-text and semantic search work alongside a versioned case briefing.
- 3
Substantiate
Inspect cited documents; an internal guidance report ranks where the team should look first — before a question is asked.
Where it is used
NEXUS directs the review team to relevant documents and connections. A risk indicator alone is not proof of misconduct; interpretation, validation and the final assessment remain with authorised professionals.
Procurement and supplier fraud
Related-party suppliers, prices above market, purchase orders raised after the invoice, bank details matching an employee.
Invoice manipulation
Duplicate numbers, arithmetic that does not add up, invoices in correspondence but absent from the ledger.
Conflict of interest and leakage
Private-address correspondence, confidential documents forwarded before a transaction, recipients removed from a chain.
Regulatory and audit response
Complete coverage, item-level citation and an append-only ledger — designed for the exchange with a regulator.
Built for: Internal audit and compliance · Forensic practices · Banks · Law firms
Use in practice
- live corporate investigations
- 4
- items across current cases
- ≈1.3 million
- items in the largest case, on one workstation
- 692,626
- fabricated answers in tested sets
- 0
Internal, case-specific results — not a general accuracy guarantee.
Security and technical details
Evidence integrity at the core
Custody
SHA-256 on intake, a sealed copy, immutable originals and a reconciliation between what was received and what was examined.
The case ledger
Every manual intervention — the action, the reason, the operator, the time — appended and never rewritten.
No number from the model
Every figure in an answer comes from a recorded query, stored with the answer.
Document text is data
Instructions inside evidence are never executed. A document that says "ignore previous instructions" is an exhibit, not a command.
Search and interpretation limits
The limits of keywords
Different wording, spelling and Turkish characters can cause relevant documents to be missed.
Where data is processed
Cloud-based tools require a separate assessment of data location, access permissions and the organisation’s confidentiality requirements.
The need to verify sources
A fluent AI answer is not a finding on its own. Each claim needs to be checked against the relevant document.
Plan a scoped evaluation using an authorised data set. Together, examine the sources behind answers, information that could not be found and the points requiring expert review.