README.md

---
title: TableMind AI
emoji: 📊
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 6.20.0
app_file: app.py
pinned: false
license: mit
short_description: Fine-tuned table Q&A chatbot with RAG and auto-charts
---

# TableMind AI

Upload an Excel, CSV, PDF, or Word document (any size) and ask questions in plain English.
Numeric/aggregate questions are answered with exact computed results via a pandas execution
layer, not guesses — everything else is answered via retrieval-augmented generation over the
document. Toggle the chart option for a visual on comparison-style questions.

Built with a QLoRA fine-tuned `Qwen/Qwen2.5-1.5B-Instruct`, `sentence-transformers` + `FAISS`
for retrieval, and a text-to-pandas router for numeric accuracy.

> **Hardware: CPU basic.** This version runs entirely on CPU — no ZeroGPU dependency, no
> daily GPU quota, no PRO subscription required. Generation is slower than GPU hardware
> (expect several seconds to tens of seconds per answer depending on question length), but
> it's free and avoids ZeroGPU's worker-allocation/quota issues entirely.