AI Business Intelligence & Automation Platform
Turn business data into decisions.
Pulse connects your business data, transforms it into clear intelligence, and gives you an AI-powered command center for understanding what happens next.
No sign-up. The demo opens an isolated session with a fictional company, Northstar Commerce — 25,000+ customers, 65,000+ orders and 550,000+ events of synthetic data.
The problem
Data everywhere, answers nowhere.
Scattered and messy. Customers live in a CRM export, orders in the storefront, payments in a processor report — with duplicates, mixed date formats and five spellings of “United States”.
Dashboards that don’t explain. A chart shows revenue fell. It doesn’t say which category, channel or refund pattern caused it.
AI that guesses. Pasting spreadsheets into a chatbot produces confident, unverifiable numbers.
Insights that go nowhere. Findings need to become follow-ups, escalations and recovery tasks — automatically.
How it works
From raw exports to automated action, in one system.
- 01
Messy data
CRM, storefront, payments & supplier exports
- 02
Ingestion
CSV upload, column detection & mapping
- 03
Cleaning
normalize, de-duplicate, link, reject
- 04
Database
relational Postgres with Row Level Security
- 05
Analytics
every metric computed in SQL
- 06
AI interpretation
explains data retrieved by tools
- 07
Action
insights → automations → tasks
Live dashboard
Revenue, orders, AOV, conversion, retention, refunds and margin — with cohort retention, funnels and channel breakdowns. Every figure is a SQL aggregation.
Data Quality Center
Scan 17,000+ raw records from four source systems, see every issue, then clean them: normalization, duplicate merging, relationship matching and a before/after quality score.
Business insights
Generic detectors surface category declines, refund spikes, payment-failure incidents, fast-growing regions and high-value customers going quiet — with the numbers behind them.
AI assistant
Ask in plain language. The model chooses from 17 read-only analytics tools, then separates facts from interpretation. It never sees your whole database or writes SQL.
Automations
In-database rules turn findings into de-duplicated actions and notifications: win-back follow-ups, payment recovery, ticket escalations, refund investigations.
Reports & follow-ups
Monthly executive reports computed from SQL with a clearly labelled AI narrative, and personalized customer e-mails grounded in each customer’s real history.
AI that discovers, not invents
Grounded answers, visible sources.
The dataset contains real business stories — nothing about them is written into the prompts. The analytics layer measures them; the AI explains what the measurements show.
- Numbers rendered as cards and tables come straight from query results
- Facts, interpretation and recommendations are labelled separately
- If the data can’t answer: “I don’t have enough data to answer that accurately.”
- Graceful fallback when the AI provider is unavailable
Why did revenue fall this month?
Tools called
get_revenue_summaryget_category_performanceget_top_productsget_channel_performanceThe assistant compares the two periods, finds which categories and products moved, checks refunds and channels — and answers with the exact figures it retrieved.
Try it live in the demo — answers use the synthetic dataset.
Security
A public demo that is still private by design.
Row Level Security
Multi-tenant Postgres. Every table is protected in the database; demo visitors can read the demo organization and write only rows tagged with their own session.
Anonymous sessions
Visitors get an anonymous Supabase session — no sign-up — isolated from every other visitor. Reset or exit at any time.
Server-only AI
The OpenRouter key never reaches the browser. The model can only call fixed, validated, read-only tools; there is no path to arbitrary SQL.
Explore the working system.
Next.js · Supabase Postgres & Auth · OpenRouter · deployed on Vercel. Built as a portfolio demonstration.