Skip to content
All projects
  • Web Applications
  • Machine Learning / AI
  • Business Intelligence

InsightAI — AI Business Intelligence Analyst

Ask your sales data questions in plain English. Get answers whose numbers are checked.

A full-stack platform where small businesses upload sales data and ask questions in plain English. An AI analyst plans each analysis, a Python engine performs every calculation, and answers return with charts, forecasts and explanations whose numbers are verified against the engine's results.

  • TypeScript
  • Python
  • Next.js
  • React
  • FastAPI
  • PostgreSQL
  • Pinecone
  • OpenAI API
  • RAG
  • pandas
  • Docker
  • Stripe

01

Project overview

What it does
A self-serve BI analyst for small businesses: upload sales data, ask a question, get a chart, a forecast and a plain-English explanation.
Why I built it
LLMs are good at understanding questions and explaining results, and unreliable at arithmetic. InsightAI splits the job so each part does what it is good at.
What it solves
Small businesses have the data but not the analyst. Generic chatbots answer confidently with numbers they made up.
Who it's for
Small-business owners and operators without a dedicated data team.

02

The problem

Small businesses sit on sales exports they rarely analyze, because BI tools assume an analyst and chat assistants hallucinate figures. A trustworthy assistant has to understand loosely worded questions without ever being the source of the numbers.

03

The solution

The AI analyst (OpenAI API) interprets the question and produces an analysis plan, grounded with retrieval over the user's context via Pinecone. A Python engine built on pandas executes the plan and computes every figure. The written answer is then checked against the engine's results before it is shown, alongside charts and forecasts. A FastAPI backend and PostgreSQL store users, datasets and history; the Next.js / React frontend handles upload and conversation; Stripe handles billing; Docker packages the services; Playwright covers end-to-end flows.

04

Architecture & workflow

  1. STEP 01

    Upload

    Sales data from the business

    • Next.js
  2. STEP 02

    API

    Auth, datasets, sessions

    • FastAPI
    • PostgreSQL
  3. STEP 03

    Plan

    AI analyst plans the analysis

    • OpenAI API
    • Pinecone RAG
  4. STEP 04

    Compute

    Python engine does every calculation

    • pandas
  5. STEP 05

    Verify

    Answer numbers checked against engine output

  6. STEP 06

    Answer

    Charts, forecasts, explanation

    • React
Data flow · source to decision

05

Technology stack

Languages

  • TypeScript
  • Python

Frameworks

  • Next.js
  • React
  • FastAPI

Databases

  • PostgreSQL
  • Pinecone

AI / LLM

  • OpenAI API
  • RAG

Data Engineering

  • pandas

Cloud & Infrastructure

  • Docker
  • Stripe

Testing

  • Playwright

06

Project demonstration

Uploading a dataset and asking three questions end to end.

07

Key features

Plain-English questions

No SQL or pivot tables — ask the way you'd ask an analyst.

Verified numbers

Every figure in the answer is checked against the computation engine's results.

Charts and forecasts

Answers come back visual, with forward-looking forecasts where relevant.

Context-aware retrieval

RAG over Pinecone grounds the analyst in the business's own context.

Production SaaS shape

Accounts, persistence, Stripe billing and Dockerized services.

End-to-end tested

Playwright tests cover the critical user flows.