Skip to content

WORKFLOW.md

Latest commit

 

History

History
52 lines (42 loc) · 1.52 KB

WORKFLOW.md

File metadata and controls

52 lines (42 loc) · 1.52 KB
Raw

DataIntel AI — Workflow

Problem

Non-technical users can't extract insights from CSV data without SQL/Python knowledge.

Solution

Upload any CSV → ask questions in plain English → get verified, data-backed answers with auto-generated charts.

How It Works (2 LLM Calls Per Query)

CSV Upload → Pandas ingests → SQLite on disk → DataFrame deleted from RAM
                                    │
             ┌──────────────────────┘
             ▼
   Gemini Call #1: Schema + Question → SQL Query
             │
             ▼
   sqlite3 executes SQL on real data (anti-hallucination)
             │
             ▼
   Gemini Call #2: SQL Results → Answer + Explanation + Chart JSON
             │
             ▼
   Frontend renders response + Recharts visualization
             │
             ▼
   SQLite DB deleted (privacy-safe, zero persistence)

Tech Stack

Layer Tech
Frontend React 19, Vite, Tailwind CSS v4
Charts Recharts
Backend Python FastAPI
Database SQLite (on-disk, temporary)
AI Google Gemini 2.5 Flash
Export CSV, Excel (OpenPyXL), PDF (ReportLab)

Key Differentiators

  • Zero Hallucination — SQL executed on real data, not LLM imagination
  • Scalable — SQLite on disk (~5MB RAM), not Pandas in-memory (~500MB+)
  • Only 2 API calls per query — optimized for free-tier limits
  • Full Transparency — exact SQL shown to user for every answer