Document intelligence and financial analysis for banks, funds and advisors.
Amaril reads, reconciles and reasons over financial documents, filings, reports, contracts, and answers with cited sources, financial ontologies and generated memos. It's built for teams where a wrong number has consequences.
Grounded by design
Every answer traces back to the document it came from. Financial ontologies give the numbers structure, so Amaril reasons over entities and relationships, not over similar-looking text.
The difference shows on the cases that matter. Searching by textual similarity means treating two line items with nearly the same name as interchangeable, even when they belong to different periods, entities or currencies. An ontology prevents that confusion because it knows what an entity is, which period it belongs to and how it relates to the others. In finance this isn't an academic refinement: it is the difference between a correct number and a plausible one.
Where a wrong number has consequences
People working on financial statements, reports and contracts aren't looking for a tool that writes well: they're looking for one they can rely on when someone asks where a figure came from. That's why Amaril is built around three non-negotiable behaviours:
- It always cites. Every claim carries the document and the exact passage behind it, so checking takes seconds rather than an hour of searching.
- It reconciles rather than summarises. When the same figure appears across several documents, the comparison is the work: discrepancies must be surfaced, not smoothed away into an elegant synthesis.
- It abstains. When the sources don't support an answer, saying so is the correct answer. A system that cannot say "not found" has no place in a regulated context.
The memo is the finish line, not the starting point
Generating text is the easy part. The real work happens before: reading the right sources, reconciling figures across documents and reconstructing the relationships between entities. When that part is solid, the memo is almost a consequence — and it is grounded, line by line, in what the documents actually say.
Built alongside the people who use it
Amaril comes out of the same method we bring to client projects: watch how a task is really performed before deciding what to automate. It isn't a generic model with a financial interface on top, but a product shaped around the documents, constraints and responsibilities of people working in banks, funds and advisory firms.
It is also why we run it in production ourselves. A product we use and maintain every day tells us where the work actually jams, long before any support ticket would.