MILabs
Integrations

Document Q&A assistant with FYI: how the build works

How a Document Q&A assistant works with FYI, what it answers from your own documents, and the steps to plan the build.

Updated 2026-09-16

Document Q&A assistant with FYI

Firms that keep client records in FYI often face the same delay with documents. A routine question, such as which clause sets a notice period, can depend on a file only one person knows how to find. A Document Q&A assistant reduces that search. It answers questions from a fixed set of documents the firm chooses, locates the passages that bear on each question, and cites them so a staff member can check the wording. The first candidates are usually the files staff consult most: lender guides, engagement letters, internal memos and standard terms. No connector to FYI is confirmed, so the document set would likely be gathered by staff and passed in through a checked route. Client data and access rights should be settled before anything is loaded. A short list of real team questions and agreed limits will guide the build.

How it works

  1. Define the document set with the team. List the lender guides, engagement letters, memos or standard terms the assistant may read, and leave out any file the firm cannot share with a language model.

  2. Settle access before any copying begins. Confirm which FYI records may leave the platform and whether the firm's own settings allow a checked export. Where they do not, staff can pass in copies by hand.

  3. Ingest the approved files into a controlled index. LlamaIndex can split each document into passages that keep their source references, and pgvector can store the search vectors in PostgreSQL.

  4. Connect the Anthropic API to produce replies. Each answer should be built only from the passages retrieved for that question, and the instructions should tell the model to say when the documents do not cover it.

  5. Test with questions the team actually asks. Check each cited passage against its source, and log every case where a reply goes beyond what the document says.

  6. Decide how staff use the output. Replies stay in the assistant's own screen, so anything a staff member wants on a client record is written there by hand unless a checked route to FYI is confirmed.

good questions, straight answers

Frequently asked questions

Can a Document Q&A assistant work with FYI?
Yes, through the documents. The assistant answers from a fixed set of files that the firm supplies. Those files can reach it by manual copy from FYI, or by a checked export if the firm's own settings allow one. No public API, connector or webhook is confirmed for FYI, so a live link should not be assumed. Confirm the export options before planning the build.
What does a Document Q&A assistant write to FYI?
In the basic build, nothing. The assistant writes replies to its own chat screen and keeps an index of the files it was given. It does not post notes, tasks or documents back into FYI, because no write path into the platform is confirmed. If a firm later wants summaries saved on a client record, that step should be designed separately, with staff reviewing each entry before it is saved.
What does a Document Q&A assistant with FYI need from me?
It needs a short list of the documents to cover, the questions your team actually asks, and clear rules on what the assistant should decline to answer. It also needs a decision on client data, since files from FYI may hold personal or financial details. Your team will check replies against the cited passages, so someone on staff should own that review from the start.
What does a Document Q&A assistant not do with FYI?
It does not change records, file documents or send messages on the firm's behalf. It does not act as a case management system, and it does not decide what counts as correct advice. A staff member still makes every judgement, and the assistant only points to the passages it relied on. Where the documents are silent, the reply should say so rather than guess.
Which tools sit between a Document Q&A assistant and FYI?
Between the files and the chat screen sit a few components. LlamaIndex splits each document into searchable passages and keeps their source references. pgvector, an extension for PostgreSQL, stores the vectors used for search. The Anthropic API produces the written reply from the passages retrieved. FYI itself sits outside this chain, since only exported or copied files reach the index.
let's make something happen

Custom development: how we can help

Choose how you want to work on this with MILabs. Both options start with an enquiry.

Discovery is free, no obligation from the first chat.

Let us build it for you

We build it and look after it.

Build it yourself, we'll help

You build it with our direction, teaching and a weekly check-in.