Canonical fund data, CCI Product Summaries, PRIIPs KIDs, and SFTP delivery, built as regulated workflow. Deterministic core; bounded LLMs only where they belong; field-level audit trail on every output.
Regulated output cannot drift between runs. Helix executes the transformations that touch a published document as versioned, unit-tested code. Language models are confined to the perimeter, where ambiguity actually lives (document extraction, free-text classification), and their output is schema-validated before it enters the workflow.
PRIIPs and CCI calculations execute as standard. Tested by regulators, trusted by clients.
AI handles the heavy lifting, in the places it works. Humans in the loop. Always.
Every file is tracked, every change is logged. Every operator, document, timestamp, data field. Replayable on demand, three weeks or three years later.
Helix CCI is the first product on the platform. It ingests fund data, extracts and validates against schema, manifests the products you intend to publish, and renders the regulated document. Every cell on every screen below points back to a source.
Drop CSV or XLSX. Column types are auto-detected against the canonical schema; mismatched columns are flagged before the run starts.
Each file is validated before any record reaches the canonical store. Malformed rows are quarantined and surfaced for review, never silently coerced.
Multiple files per upload. The run links them by ISIN and as-of date, so a single share class never assembles from data that disagrees with itself.
Every run is logged: file, data type, row count, ISINs covered, rules applied.
Flagged records pause the run. They are reviewed, never overridden in the background.
LLMs read and classify. They never write to the canonical record. Every output is checked against the schema first.
Products are addressable by ISIN. Each carries a status (Draft, Generated), the manufacturer LEI, and the transform versions used.
Generate is a four-eyes operation; Draft means data is canonical but a human has not signed off.
Export CIDs emits the CCI exchange format for distributor ingestion. No screen-scraping, no PDF re-parsing.
The CCI Product Summary, created from the verified data. Every field is traceable to its source file, transform version, and approver. Branding (forest green here) is configured per manufacturer, along with layout, formatting, fonts and styles. Your document, not a boilerplate output.
Helix is composed of four products that share a canonical fund record. Adopt one; the others slot in as your distribution obligations expand.
A single typed record per share class, versioned and immutable. Reconciled from custodians, fund accountants and transfer agents. Schema-validated against FundsXML 4.2.x, EMT v4.3, and openfunds CDS v2.13.0 on every write.
Consumer Composite Investments templates rendered from the canonical record. Generated, not transcribed. Multi-language; trackable to the field; bookable into a four-eyes approval workflow.
RTS 2017/653 compliant. SRI, performance scenarios, costs grid produced by deterministic functions. Prior runs reproduce to the field; corrections are diffable against the published version.
Recipient lists are pinned, payloads are RSA-signed, delivery receipts persisted. No portals, no scraping; the format your counterparties already ingest. Failed deliveries open a ticket, not a silence.
Distribution today is held together by spreadsheets, portals, and email. The rows below set each part of that setup next to its Helix equivalent.
Audit lives in comment threads, handover notes, and the memory of whoever ran it. Replay is a manual exercise.
Every field carries its source, transform version, and approver. Replay is a query, not a project.
Versions live in filenames. Reproducing a prior quarter means rebuilding it from memory.
Every transform is versioned. Each published document carries the manifest used to produce it.
Data is coerced into shape on the way in. Format errors surface late, often at distribution.
Every record is validated against the canonical schema. Non-conforming rows are held back before they reach the store.
Each counterparty has its own portal, format, and login. Some still require manual upload.
Counterparties receive the format they ingest, signed and acknowledged. Failed deliveries open a ticket.
Each new format means a new vendor, a new integration, and a new round of testing.
New formats are added to the canonical schema. You opt in to an output, not a project.
Three to six weeks across administrator, manufacturer, and translator.
Hours. The canonical record is reused; additional languages are a render setting.
All products are built directly from regulatory guidance and best practice. Full regulatory specifications available on request.
Governs the SRI, performance scenarios and costs disclosed in a KID. Implemented as a versioned set of deterministic functions, tested against ESMA worked examples.
The new Consumer Composite Investments format that will replace UK PRIIPs. Helix renders both today; the canonical record covers either output without re-keying.
The European MiFID Template for distributor exchange. Helix produces EMT exports as a first-class artefact, not a download from a portal.
Open standard for fund data interchange across asset managers, distributors, and data vendors. The canonical record maps directly to openfunds fields, so distributor exports require no re-keying.
Schema for inbound fund accounting and custodian data. All ingestion is schema-validated; non-conforming records are quarantined, not silently coerced.
Outputs are auditable at the field level, which supports the "monitor outcomes" pillar. Operators can answer "what did the customer see, and why" in a single query.
Helix is in production with two fund administrators and four UK asset managers. Access is gated; we onboard one new firm at a time, and the conversation begins with your engineering team.