Independent Contractor/Notebook AI

Notebook AI — AI Data Analysis Platform

A full-stack platform that combines chat, Jupyter notebooks, and file-based analysis into a single conversational interface.

Overview

Notebook AI is an AI-powered data science assistant that lets users upload datasets, ask questions in plain English, and receive real-time code execution, visualizations, and insights — without writing a single line of code. It bridges the gap between chatbots and notebooks by tightly integrating both into one seamless workflow.

The Problem

Data analysis today is fragmented across multiple tools — chatbots for Q&A, Jupyter notebooks for coding, and file explorers for data. This constant context switching slows analysis, increases cognitive load, and makes data science inaccessible to non-technical users.

Existing AI tools can explain concepts, but cannot reliably execute code on real datasets or preserve analytical state across a conversation.

Target Users

•

Founders & product managers — exploring datasets without engineering support

•

Analysts & data scientists — prototyping insights quickly

•

Non-technical users — who want answers from data, not code

•

Developers — who want an interactive AI + notebook workflow

How the Platform Works

1.Conversational Interface

Users chat with an AI model using real-time streaming responses — fast, natural, and familiar.

2.Automatic Notebook Mode

The system detects when a request involves data analysis (file upload or keywords like "analyze" or "plot") and switches to notebook mode automatically — no manual toggle required.

3.Jupyter Notebook in Chat

The AI generates Python code, executes it inside a secure cloud sandbox, and streams outputs — tables, charts, and errors — inline in the chat as notebook cells.

4.Persistent Analytical State

Variables and datasets persist across messages within a conversation, enabling multi-step analysis just like a real notebook.

5.Artifacts & Documents

The AI creates rich artifacts — documents, code files, notebooks, sheets, and images — that open in a side panel for deeper editing and exploration.

6.Side Chat for Quick Follow-ups

Users can select any output or text and ask a follow-up question in an ephemeral mini-chat without disrupting the main conversation.

Architecture & Core Components

  • •Client (Next.js App Router) — Streaming UI with a chat interface, notebook viewer, and artifact side panel — all updating in real time.
  • •AI Layer (Claude via Vercel AI SDK) — Handles streaming responses, tool calling, and multi-step reasoning using Claude Sonnet 4 and Haiku 4.5.
  • •Execution Layer (E2B Sandboxes) — User code runs exclusively inside isolated E2B cloud containers — never on the server. Supports pandas, numpy, matplotlib, seaborn, plotly, and scikit-learn.
  • •Storage Layer — PostgreSQL (Neon) for chats, messages, notebooks, and artifacts. Vercel Blob for file uploads and dataset storage.
  • •Auth & Security — JWT-based authentication with guest and registered sessions, rate limiting per user type, and Zod validation on all API inputs.

Tech Stack

Frontend

Next.js 16 (App Router + Turbopack)Next.js 16 (App Router + Turbopack)
React 19, TypeScript, Tailwind CSS, shadcn/uiReact 19, TypeScript, Tailwind CSS, shadcn/ui
Framer Motion
CodeMirror (code), ProseMirror (documents)

Backend

Next.js API RoutesNext.js API Routes
Vercel AI SDK (streaming + tool calling)
NextAuth (Auth.js)
Drizzle ORMDrizzle ORM
PostgreSQL (Neon)PostgreSQL (Neon)
Vercel Blob

AI & Execution

Claude Sonnet 4 / Haiku 4.5 (Anthropic)
E2B Code Interpreter
pandas, numpy, matplotlib, seaborn, plotly, scikit-learn

Infrastructure

Vercel (Deployment)
OpenTelemetry (Monitoring)
Playwright (E2E Testing)

My Role — End-to-End Ownership

Built entirely as a solo engineer, with full ownership across every layer:

  • •Designed the system architecture and data models from scratch
  • •Built real-time streaming chat with AI tool orchestration
  • •Implemented a Jupyter-style notebook engine with sandboxed cloud execution
  • •Designed cell versioning and persistent notebook state across conversations
  • •Built the artifact system — documents, code files, sheets, notebooks, and images
  • •Implemented auth, rate limiting, file uploads, and storage
  • •Built the complete frontend — chat UI, notebook viewer, and artifact panel
  • •Deployed and productionized the full platform on Vercel

Outcome

Notebook AI delivers a unique chat-plus-notebook experience that lets users analyze real datasets conversationally. It makes data science faster, more accessible, and more interactive — for both technical and non-technical users — without requiring any prior coding knowledge.