From raw feeds to AI-ready data for Nzuri

Overview

Portfolio data lived in spreadsheets, custodian files and third-party feeds, so every report started with manual reconciliation. We built automated extract, transform and load pipelines into a single governed data layer, then made that data AI-ready so language models answer from approved numbers and cite their sources.

• Automated ETL pipelines that extract, clean, reconcile and load portfolio data on a schedule
• AI enablement: an assistant grounded in the governed data layer, with every answer traceable to its source
• A dashboard and globe view fed from the same pipeline, tuned for large displays and tablets

Tech stack

Extract

Connectors pull positions, transactions, valuations and market data from custodian files, fund administrators and third-party feeds on a schedule, with every run logged so gaps are visible rather than silent.

Transform

Raw data lands untouched, is cleaned and reconciled into a governed model (bronze, silver, gold), and is documented as it goes. Definitions for returns, exposure and attribution live in one place instead of in a dozen spreadsheets.

Load and serve

The gold layer is served to the dashboard, exports and the AI assistant from the same source, so a number on the globe matches the number in the board pack.

AI enablement

The assistant only sees approved, documented data. It answers questions in plain language, shows the figures behind each answer, and says so when the data can't support one.

Executive UX

We optimized for scan-first layouts: headline KPIs that hold up at arm's length, then progressive paths for PMs who need lineage. Nothing decorative without a decision attached.

Impact

Reliable pipelines first, then AI that can be trusted with the numbers.

Reporting moved from manual month-end reconciliation to automated runs, and executives could ask follow-up questions that the model answers from the same governed dataset. The win is fewer stitched-together decks and faster alignment before allocator calls.