A chat assistant built into an electronic document management system takes plain-language requests, turns them into reports with charts, searches documents by meaning, and shows each user only what they’re allowed to see
The organization keeps its directives, inquiries, correspondence, and assignments in its own electronic document management system (EDMS). The archive holds up to 100,000 documents, and employees use the system in several languages.
To keep tabs on their departments, leaders need reports: how many documents have been completed, which are overdue, and how busy each department is. The data was all in the system, but only people who knew which function to call and which parameters to set — period, department, document types — could actually get a report.
Reports came in fixed formats and were slow to generate on large datasets. Search only matched exact words, so it missed documents that used synonyms or matched a query only in meaning. Reports and search lived in separate sections, and queries had to be written in the interface language.
The goal was to bring reports and search into a single chat: users type a request and get back a report or the matching documents. The core system had to stay untouched, and access rights had to be enforced.
Users describe what they need in plain language, and the system works out the report type and parameters. They can export the finished report to Excel or PDF, or open it as an interactive chart with click-through details, and save the important ones to favorites. Heavy reports build in the background, and the system emails the user when they’re ready.
The system interprets what a query means while also running a keyword search, ranks results by relevance, backs them up with quotes from the documents, and breaks complex questions into several steps.
Users can ask in any of several languages and get both reports and search in the same chat.
Before showing any results, the system checks the user’s permissions and filters out documents they can’t access.
Let’s follow one request step by step, from a question in the chat to a finished report or search results.
The user types a request into the chat in the system’s side panel — for example, “show assignment completion by department for the quarter” or “find directives on business travel.” The system decides whether it’s a report request or a document search. The chat uses the core system’s login, so there’s no separate sign-in.
A classifier module picks the right report out of 25 types, and an extractor module pulls the period, department, document types, and language from the request. The system then calls the prebuilt report function on a read-only database replica that refreshes every 15 minutes, so the production database takes no extra load.
Simple queries get a single search pass; complex ones are broken into several steps. The user then sees the top ten documents, with filters for date and status.
Every action goes into an audit log: who searched, which documents they found, and which report they requested. Users rate each report as “useful” or “not useful,” and the team uses those ratings to decide what to improve.
Also used: Docker — containerized deployment · Prometheus and Grafana — monitoring
The organization now has a chat assistant built into its EDMS: users ask a question in plain language and get back a finished report or the document they were looking for.
This works for any organization with a large document archive where only people who know the system inside out can pull reports or find documents.
We build solutions like this around your needs: turnkey AI products, AI agents and automation, integrations, and data.
Our servicesTell us about your project — we’ll talk it through and show you how to make this work for your business
We use cookies and Yandex Metrica to improve our website. Privacy policy
