If your Servoy solution has documents behind it, you have a steady stream of questions that land in someone’s inbox. Where is the current HR leave policy? Which version of the compliance standard are we on? Does the contract template cover this clause? Each one means someone stops what they are doing to search a shared drive, dig through a folder, skim a PDF, and paste a link back. Meanwhile the person who asked has already sent a follow-up.
We have a document assistant that ingests Markdown, HTML, PDF, and Word files, answers natural language questions in 2-3 seconds with sourced links to the original documents, and is designed to bolt right onto a Servoy solution.
Type a question. Get a sourced answer with clickable links to the exact documents that support it, rendered natively in the browser. The retrieval pipeline (hybrid vector and keyword search, plus cross-encoder re-ranking) finishes in seconds.
Full architecture and screenshots: RAG Document Assistant: From Single-Purpose Chatbot to Multi-Repository Document Platform - Dotzlaw Team
What it does:
- Point it at a folder, ingest into a repository: Each folder of documents becomes its own repository (HR Policies, Legal Contracts, Compliance Standards, Technical Docs, whatever you need). Add a new repo by pointing the ingester at a new folder.
- Per-repository user access control: Repositories are filtered at query time based on which repos a user is allowed to see. Add authentication and per-user repo permissions without restructuring the index.
- Search one repo or across all of them: A user can scope a query to a single repository or search across every repo they have access to in a single call.
- Four file formats, one ingestion pipeline: Markdown, HTML, PDF, and Word documents all flow through the same parse-chunk-embed-index pipeline.
- Hybrid search with cross-encoder re-ranking: Vector search catches semantic matches. BM25 keyword search catches exact terms like part numbers, regulation codes, or function names. A cross-encoder re-ranks the top candidates against the actual query for relevance.
- Source attribution on every answer: Every response includes clickable links to the documents it drew from. Click a source, the original document opens in the viewer (PDF as PDF, DOCX rendered as PDF, HTML and Markdown with dark mode support).
- Admin interface: Ingest folders, manage documents (delete, re-ingest), create and delete repositories, and view analytics including a gap analysis that surfaces questions users asked but the corpus could not answer.
Why it embeds cleanly into a Servoy solution:
The viewer is iframe-based and format-agnostic. Whatever can render in a browser iframe (which is everything the system serves) drops into a Servoy solution without custom frontend work. The same architecture means it also runs as a standalone application, or embeds into a company intranet, internal portal, or any other web application that can host an iframe.
Use cases we have built or scoped:
- Internal HR policy lookup, scoped to employees only
- Compliance and regulatory search across hundreds of NIST or ISO PDFs
- Legal contract template search across a DOCX library, scoped to the legal team
- Technical documentation search across HTML doc trees
- Customer-facing knowledge base for a SaaS product, with separate internal and external repositories
Read-only access to the source documents is all the system needs. Your existing applications and data are untouched.
If you want a live demo, want to talk architecture, or want to see what this looks like running against your own document set, reach out. The Dotzlaw team can help get it integrated into your Servoy solution quickly, or stand it up as its own application against the documents you already have.