AI Document Explainer

Upload a document and get an instant summary, key points, action items, and a chat interface to ask follow-up questions — all grounded in the document's own content, powered by Google Gemini.

Live at ai-document-explainer-mu.vercel.app.

Screenshots

Upload — drag and drop or browse for a file, with a live cost/size hint before you commit.

Analysis — the original file previews inline (native rendering for PDF/image/audio) alongside the AI-generated summary.

Chat — key points and action items reveal with a highlighter-style sweep; ask follow-up questions grounded only in the document's content.

Light mode — the same "paper and ink" palette, by daylight instead of lamplight.

Features

• Broad format support — PDF, PNG, JPG/JPEG, audio (WAV, MP3, AAC, OGG, FLAC, M4A), and Word/PowerPoint (.docx, .pptx). Audio and images are understood natively by Gemini; Office documents are text-extracted server-side first (Gemini doesn't read those formats natively).
• Structured analysis — an executive summary, key points, action items, and AI-suggested follow-up questions, produced in a single model call.
• Streaming chat — ask questions about the document; answers stream in token-by-token and are grounded only in the document's extracted content.
• Selectable model and temperature — pick between Flash-Lite (cheapest, default), Flash, and Pro tiers, and adjust generation temperature, from a settings panel. A cost/size hint shows before you commit to analyzing.
• Inline document preview — PDFs and images render inline, audio gets a native player, and Office files get a download link (no inline preview available for those formats).
• Session persistence — your analysis and chat history survive a page reload; a "Start over" action resets without needing to reload.
• Retry without re-upload — transient failures (rate limits, empty AI responses) let you retry the same upload without sending the file again.
• Light/dark theme — toggle with a default of dark.

Tech stack

• Next.js 16 (App Router, Turbopack) + React 19 + TypeScript
• Tailwind CSS v4 with shadcn/ui (base-nova style) on Base UI primitives
• Google Gemini API via @google/genai
• Vercel Blob for temporary upload storage
• officeparser for .docx/.pptx text extraction
• next-themes for theme switching
• Fraunces, IBM Plex Sans, and IBM Plex Mono via next/font/google

Getting started

Prerequisites

• Bun
• A Gemini API key
• A Vercel Blob store (BLOB_READ_WRITE_TOKEN)

Setup

bun install
cp .env.example .env.local

Fill in .env.local:

GEMINI_API_KEY=       # from https://aistudio.google.com/apikey
BLOB_READ_WRITE_TOKEN=  # from your Vercel project's Blob store settings

Run the dev server

bun dev

Open http://localhost:3000.

Scripts

• bun dev — start the development server
• bun run build — build for production
• bun start — run the production build
• bun run lint — lint the codebase

Deployment

Deployed on Vercel. Both GEMINI_API_KEY and BLOB_READ_WRITE_TOKEN need to be set on the Vercel project (Production, Preview, and Development environments) — they aren't picked up from a local .env file at build time.