A fund-data AI assistant for a South African asset manager

Fund knowledge locked in PDFs

An asset manager's most-asked questions — performance, fees, holdings, fund positioning — live in monthly fact sheets (minimum disclosure documents) and scattered market data. Answering them means a person finding the right PDF, the right page, the right month. Documents change layout without warning, and stale answers erode client trust.

An assistant that reads the documents so people don't have to

• Automated fact-sheet ingestion — monthly fund documents parsed automatically, including surviving a mid-year redesign of the entire document layout.
• Cloud data lakehouse — fund data staged through bronze/silver/gold layers on AWS, refreshed on schedule with monitoring and alarms.
• Market-data connectors — central-bank statistics and stock-exchange announcements feeding the same assistant.
• Grounded, streaming answers — responses cite the underlying fund data and stream in real time; hosted in AWS's Cape Town region for local performance.
• Run as a service — scheduled updates, alerting and an operations playbook, not a handover-and-goodbye project.

The knowledge-assistant pattern, proven on real documents

This engagement proves the document-and-data assistant pattern end to end: messy real-world PDFs in, governed data pipeline in the middle, a trustworthy conversational interface out. The same architecture serves any business whose answers live in documents — funds, insurance, legal, compliance.