Monitor changing sources
Watch sources including ČNB, EBA, and EUR-Lex; analyze new content with AI and ingest relevant material.
Applied AI · Knowledge retrieval · Monitoring
Connecting regulatory source monitoring, AI-assisted triage, and answers drawn from an indexed knowledge base.
The walkthrough below shows only the monitoring module of a broader compliance application. The platform combines a RAG knowledge base with citation verification, relevance-weighted retrieval and a Model Context Protocol (MCP) server. In its knowledge-base mode, the chatbot answers exclusively from relevant stored information, respects user access permissions, and maintains logs and audit records.
A separate research mode can use online sources, fetching and querying external material in response to the question. The application also reviews documents, including internal policies, to assess alignment with rules and regulations and identify potential risks for compliance teams.
Users can choose and switch between supported OpenAI and Anthropic models.
Regulatory knowledge spans multiple sources and keeps changing. This project connects discovery of new material with a question-answering workflow for Czech and EU banking-compliance topics.
I began as the business analyst, gathering stakeholder needs and requirements and translating them into user stories.
I then designed the application architecture. For the MVP, I combined third-party services and components, including Qdrant, Supabase and Redis-backed RQ workers for background processing.
I also designed the frontend and built the full application through AI-assisted coding, covering the interface, backend and integrations.
Watch sources including ČNB, EBA, and EUR-Lex; analyze new content with AI and ingest relevant material.
Combine retrieval with citation checks and refusal behavior when evidence is missing.
Separate implementation checks from answer-quality evaluation. A matching citation alone does not establish a correct or complete answer.
The applications are deployed and remain in active development. Work on the next phase is underway, focused on answer-quality evaluation, support for scanned PDFs, improved retrieval ranking and broader source coverage.