Compliance work, provable.
Internal investigations, regulator-facing reviews, audit findings — Felarity produces a signed, verifiable record of the analysis. The work product holds up after the meeting ends, after a witness changes their story, and after the regulator asks how you reached your conclusion.
What finance teams use Felarity for
Finance compliance work runs on conversations: interviews, walkthroughs, attestations, and the meetings where someone says something that does not quite line up with what they said three weeks ago. Felarity is built for the moments where that mismatch matters and needs to be defensible later.
- FRB / OCC consent-order remediation. Defensible review of recorded conversations tied to the consent order's required actions. Each transcript and each contradiction is sealed to the source audio and to the council's reasoning, so the remediation file does not depend on a junior analyst's recollection of what was said in week three.
- Internal audit interviews. Surface inconsistencies across multiple sittings with the same control owner — the answer in the first walkthrough versus the answer in the follow-up versus the answer to the audit committee. Felarity tracks who said what, when, and how the answer drifted.
- SOX 404 control walkthroughs. Track who claimed ownership of which control, where the boundary sits between two teams, and whether the same person assigned that control to someone else in a different walkthrough. The control matrix and the walkthrough record stop being two separate documents you have to reconcile by hand.
- Whistleblower intake review. Confrontation classification and acoustic credibility markers are surfaced for human review, never as a verdict. The investigator decides; Felarity provides the structured evidence and the chain-of-custody record that makes the decision reviewable later.
The Finance council
When you run a meeting or an interview under the Finance industry profile, four specialist analysts review the transcript in parallel before synthesis. Each one looks at the same conversation through a different lens, and disagreements between them are flagged rather than averaged away.
Reads for financial substance. Flags numbers that do not reconcile across statements, revenue-recognition language that does not match the underlying transaction, and forecast claims that drift from prior guidance.
Reads for the control angle. Identifies who owns which control, where ownership is being quietly reassigned, and where the conversation describes a control operating differently than it is documented.
Reads for process. Tracks walkthrough completeness, evidence-gathering claims, and statements about management-review controls — particularly the ones that depend on someone having actually read the report they signed.
Reads for regulatory posture. Flags statements that could create disclosure obligations, language that conflicts with prior public filings, and admissions that need to be routed to outside counsel before they propagate further.
Working with regulators
The 8-node SHA-256 Merkle attestation chain is built to support regulator-facing files. The chain root and each node are signed with Ed25519. The regulator's own technical team can verify the chain against our published public key without a Felarity account, without our infrastructure being reachable, and without trusting any claim we make about the work product. Bring the chain JSON and the public key — that is the entire dependency. Verification instructions live in the audit chain guide.
MNPI and information barriers
Workspace isolation enforces information-barrier separation at the data layer: a workspace cannot read another workspace's transcripts, contradictions, council outputs, or speaker samples, and admin role assignments are scoped per workspace rather than per account. The workspace audit log records every access — who opened which session, who exported which report, who ran a re-analysis and against what.
If you run a public-side / private-side split, provision separate workspaces and assign admins independently. The audit log is designed to satisfy a regulator who asks the obvious next question: show me everyone who saw this transcript.
Workflow example
A typical consent-order remediation interview, end to end:
- Compliance investigator opens a Felarity session under the Finance profile. Microphone consent is captured and timestamped. Workspace is the consent-order workspace, not the general compliance workspace.
- Live transcription begins. The council runs in the background on a deep model, flagging contradictions against prior interview transcripts already in the workspace memory.
- An intel card surfaces mid-interview: the subject's answer about who approved the control change conflicts with what they said in the walkthrough two weeks ago. The investigator notes it for follow-up rather than confronting in the moment.
- Interview ends. The investigator stops the session and the 11-stage post-session pipeline runs: full-session diarization, acoustic markers, confrontation classification, speaker attribution of contradictions, DeBERTa-v3 NLI re-scoring, topology analysis, deep council re-review with speaker context, and the Merkle attestation chain.
- Investigator reviews the final report. The contradiction is now attributed to the named speaker, scored, cross-referenced against the prior transcript, and sealed to the source audio segment.
- Report is exported to the consent-order file with the chain JSON and the verification command. The regulator's examiner verifies the chain on their own machine, against our public key, before they read the substance.
What this is not. Felarity does not produce regulatory determinations or assign culpability. The council surfaces contradictions, the pipeline surfaces acoustic and behavioral markers, and the chain proves the record was not altered. Every decision about what to escalate, what to disclose, and what to remediate stays with your compliance, audit, and legal teams.