Skip to content
AI

Privacy-Safe Synthetic Dataset Maker

Generate realistic fake data with the same shape and stats as your real data.

Intermediate16h1–2 people16/20 fit
TL;DR

Build Privacy-Safe Synthetic Dataset Maker for developers and analysts with sensitive data in ~16h with 1–2 people. Stack: Python + DuckDB + LLM + Charts.

Why it wins: Every hackathon team needs demo data, so the use case is obvious.

Build flow

01 · Problem
Teams cannot share production data for demos or tests because of privacy rules.
02 · Solution
Profile column types and distributions, then generate synthetic rows that match the stats without copying any real record, with a similarity check.
03 · Build
  1. 1Profile a CSV with DuckDB
  2. 2Fit simple per-column distributions
  3. 3Generate rows and optional LLM text fields
  4. 4Check nearest-neighbour distance to real rows
04 · Demo
Show the real and synthetic column charts overlapping while no row matches.

MVP checklist

  • Schema profiling
  • Distribution-matched rows
  • Leak check
  • CSV export

Tech stack

TechRole
PythonBackend / scripting
DuckDBIn-process analytics SQL
LLMLanguage model
ChartsData visualisation

Reference repos

Open-source projects you can build on. Links checked 2026-10-31.

Related ideas

More AI

  • AI

    Ask Your CSV in Plain English

    Upload a CSV, ask a question and get a chart plus the SQL that produced it

    Intermediate14h1–2
    DuckDBLLMCharts
    18/20

  • AI

    PDF Tables to Clean CSV

    Extract tables from messy PDFs into clean, checked CSV files

    Intermediate14h1–2
    MarkItDownPythonLLM
    16/20