Outliers Analytics

A full-stack BI dashboard over six independent business datasets — Sales & E-commerce, Customers, Marketing, Inventory, Banking & Financial, and HR — built with Next.js, DuckDB, and Recharts. Every metric is computed server-side from the source CSVs; nothing is hardcoded on the frontend.

Live demo: https://analytics-dashboard-phi-kohl.vercel.app Case study: docs/CASE_STUDY.md — overview, architecture, key decisions, and screenshots

Architecture

CSV files → Ingestion/validation → DuckDB (in-memory) → SQL aggregation → API routes → React UI

• Next.js 16 (App Router) + TypeScript — pages under src/app/, API routes under src/app/api/.
• DuckDB (@duckdb/node-api) — all analytical queries run in SQL against an in-memory DuckDB instance loaded from the CSVs at startup.
• Recharts + shadcn/ui (base-ui) + Tailwind CSS v4 — charts and components, themed via CSS variables.
• Automatic refresh — a manual "Refresh Data" button plus a 30s poll detect source-file changes (mtime/size) and re-ingest only when something actually changed.

No cross-dataset joins

Three datasets share a Customer_ID-shaped column and two share a Product_ID-shaped column, which looks like a foreign-key relationship. It isn't: joining on Customer_ID and comparing Age agrees only 1.5–1.7% of the time across every pair (random-chance level), and joining on Product_ID and comparing Category mismatches 89% of the time. These are independently-generated synthetic ID pools that happen to share a numeric range and string format.

Every page in this app is built on exactly one dataset. The Data Explorer documents this naming coincidence as a relationship diagram, explicitly labeled as unverified — see docs/adr/0001-no-cross-dataset-joins.md for the full write-up and match-rate evidence.

Pages

Page	Source dataset
Executive Overview	all six, reported per-domain (no blended metrics)
Sales & E-commerce	sales_ecommerce_data
Customers	customer_demographics_data
Marketing	marketing_campaigns_data
Inventory	product_inventory_data
Banking & Financial	banking_financial_data
HR / Employees	employee_hr_data
Data Explorer	all six, browsed independently — search, filter, sort, paginate, export CSV

Screenshots

Executive Overview

Sales & E-commerce

Customers

Marketing

Inventory

Banking & Financial

HR / Employees

Data Explorer

Getting started

npm install
npm run dev

Open http://localhost:3000. The six source CSVs in the project root are ingested into DuckDB automatically on first request.

npm run lint    # ESLint
npm run build   # production build

Project docs

• CONTEXT.md — data model glossary (Domain, Scoped identifier, Relationship candidate, Snapshot field) and dataset grain/row counts.
• docs/adr/0001-no-cross-dataset-joins.md — the architectural decision record behind the no-joins rule.