Layout-aware OCR
Extracts text from scanned financials, prospectuses, financial tables and term sheets, while preserving the document's structure. An income-statement table stays a table, a term sheet stays a term sheet.
Technology
The AI analyst that doesn't hallucinate.
Amaril builds an operational model of your portfolio: it recognises issuers, counterparties and instruments, maps relations, generates answers grounded in your financials, filings and contracts with precise citations. No opaque generation, no invented numbers.
BIRD'S-EYE VIEW
An operational model of your portfolio.
Amaril connects to your source systems and data rooms, builds an operational model that recognises issuers, counterparties, instruments and covenants and the relations between them, and returns answers grounded in your financials, filings and contracts with precise citations.
HOW IT WORKS
Four technical foundations, one promise.
Entity recognition, relation mapping, precise citations, grounded generation. Four pieces working together, so every number and every clause is verifiable at the source.
01 · ENTITIES
Every issuer, counterparty, instrument is one unified entity.
The same company may appear three ways across your corpus: full legal name in an IFRS annual report, ticker in a research report, LEI in a term sheet. Amaril recognises all three are the same subject and merges them into a single entity. Instead of isolated filings, you get a working graph of issuers, counterparties, instruments and covenants, built from your own documents.
02 · RELATIONS
Financial ontologies, rebuilt from your documents.
These are the same architectural foundations used by the operational intelligence platforms running inside the world's most demanding financial institutions. An issuer links to its filings, filings to clauses, clauses to covenants, covenants to deal deadlines. A live network of relations, not a flat archive. Queries traverse the network: ask for one issuer, get the full exposure.
03 · CITATION
Every answer reports document, page and paragraph.
No opaque generation. No invented numbers. Every figure and every clause Amaril returns is anchored to a specific passage in a real document: file name, page, paragraph. Investment-committee verification is one click away. If the source doesn't exist, the claim isn't generated. That's how hallucination risk gets eliminated at the root.
04 · RAG
Retrieval-Augmented Generation, end-to-end.
When a question comes in, Amaril runs a hybrid search: semantic (over vector representations of the filings and contracts corpus) and full-text (over exact figures, tickers and financial terminology). Results then go through a contextual reranking that scores them against the question. Only the most relevant passages become context for the generation step, which produces the final answer with automatic citation extraction.
CAPABILITIES
Four functions that work as one.
These are the pieces that make the operational deal model possible. You don't pick them individually: they run as a pipeline, each one feeding the next.
Multilingual embedding
Vector representations of your filings and contracts corpus across Italian, English, French, German and Spanish. Questions and documents can speak different languages: semantic retrieval still works.
Contextual reranking
Reorders search results by relevance to the specific context of the analysis question. A dedicated model that weighs nuance, not a static score.
NER + entity linking
Recognises issuers, LEIs, tickers, amounts, multiples and dates, and links them to the right entity in the graph. The bridge between free text in filings and a structured operational model.
Built today, designed for the years to come.
Numbers that describe how it works, not a benchmark. They're the reason Amaril holds up under real deal-room load.
Millions
of filing and data-room pages indexable in a single operational model
Every
answer cites document, page and paragraph
5
languages handled natively in semantic search
100%
of processing in EU data centers
SECURITY
The non-negotiables.
For a regulated financial institution, security isn't a list of features: it's a few architectural choices that aren't up for discussion. They're below.
Data in the European Union
All processing happens in EU data centers. No extra-EU transfer, no exposure to third-country jurisdictions. On-premise deployment also available.
Zero training on your data
Your filings, contracts and portfolio data are never used to train models. Not ours, not third parties'. Zero retention.
End-to-end encryption
Encryption at rest and in transit across the entire flow. Keys managed under European best practices.
Structural GDPR compliance
Privacy by design, native right to be forgotten, DPA available. Compliance is a property of the system, not an afterthought policy.
Complete audit trail
Every operation is logged: queries, access, exports. Immutable logs, exportable to your SIEM, ready for compliance reviews and regulators.
SSO and access segregation
SSO integration with your identity systems and segregation by deal and team. An analyst only sees the mandate documents they already have rights to access, with ethical walls and information barriers enforced.
Ready for a technical evaluation?
Our team can run an architecture deep-dive and a proof of concept on one of your real data rooms.