VOOZH about

URL: https://apify.com/salesmart-srl/polars-ai-data-transformer

⇱ Polars Ai Data Transformer Β· Apify

Pricing

Pay per event

Go to Apify Store

42 Polars AI Data Transformer

Transform datasets using natural language. Upload CSV/Excel/JSON, describe your transformation in plain English, get results + reusable Python code. Powered by AI.

Pricing

Pay per event

Rating

0.0

(0)

Developer

πŸ‘ Salesmart Srl

Salesmart Srl

Maintained by Community

Actor stats

0

Bookmarked

4

Total users

0

Monthly active users

a month ago

Last modified

Share

polars-runner

Natural-language β†’ Polars code, with auto-improving RAG cache. Runs on Apify, Docker, or as a Python library.

Part of the wiki-embedded monorepo (packages/polars-runner/).

What it does

You describe a data transformation in plain English, the runner generates Polars code with an LLM, executes it in a sandbox, and returns the results β€” plus the code itself for inspection and reuse.

  • Multi-format input: CSV, JSON, Excel, Parquet, or a JSON object passed inline.
  • Multi-table mode: ship { "contacts": [...], "companies": [...] } and the LLM gets one schema per table for explicit JOINs.
  • Three tiers:
    • BYOK β€” bring your Google / Anthropic / OpenAI / Groq key.
    • Hosted basic β€” Gemini 2.5 Flash-Lite (cheap, fast).
    • Hosted premium β€” Gemini 2.5 Pro with reasoning (slow, high quality).
  • Auto-improving: every successful execution is embedded and stored in a Pinecone index; future similar prompts retrieve and reuse working code, skipping the generation step.
  • Error recovery: on a sandbox exception, the analyzer extracts a structured hint (missing column, wrong dtype, JOIN mismatch) and asks the LLM to fix-and-retry up to N times.

Two build targets

packages/polars-runner/
β”œβ”€β”€ .actor/Dockerfile Apify deploy(apify push)
└── builds/docker/Dockerfile Standalone(docker build)

Same source, different wrappers β€” see the Dockerfile and docker-compose.yml files for runtime knobs.

Quick start

# Install (workspace member of wiki-embedded)
uv sync--package polars-runner
# Run with your own key, single table
GOOGLE_API_KEY=... polars-runner \
--input data.csv \
--prompt"Group by sector, sum revenue, top 10 desc"

Why it lives here

It is the analytics engine that backfills analytics/ chunks at compile time for ../wiki-compiler/, and may be called at runtime by ../../ for ad-hoc graph queries when a precomputed answer is not available. Standalone use is fully supported.

Status

v0.3.0 β€” refactored into the monorepo, FAANG-grade packaging. The post-execution semantic validator (planned v0.4) will tackle the "status SUCCEEDED but answer is semantically wrong" class of bugs surfaced by our benchmark.

License

MIT β€” see ../../LICENSE.

You might also like

AI Extraction Agent - Smart Scraper

alizarin_refrigerator-owner/ai-extraction-agent

AI-powered data extraction using natural language prompts. Describe what you need & let AI extract structured data from any webpage automatically.

AI Real Estate Lead Generation

actor4you/ai-real-estate-lead-generation

Generate real estate leads from Zillow using plain language queries. Describe what you're looking for in natural language and AI will find matching listings with contact information.

AI Code Reviewer - Analyze, Explain & Improve Code

ntriqpro/code-review-intelligence-mcp

Get AI-powered feedback on code quality, security issues, and improvement suggestions. Explain complex code in plain English.

XavvyNess AI Web Extractor

xavvyness/xavvyness-smart-extractor

Extract data from any website using plain English β€” no CSS selectors, no code. Describe what you want, get JSON, CSV, or Markdown back. Works even when site layouts change. Example: 'Extract job titles, company names, and salaries'.

AI Web Scraper

crawlworks/ai-web-scraper

Scrape any webpage with a URL and a plain-English prompt. Get structured JSON output powered by AI β€” no coding, no selectors, no configuration.

Smartcontext AI Web Crawler

bluelightco/smartcontext-ai-crawler

Scrape any website and extract structured data using AI-powered instructions. Provide URLs and a natural language prompt to get tailored JSON outputs.

206

5.0

Best AI Web Scraper

hgservices/Best-AI-Web-Scraper

Extract any data from any website by simply describing what you want in plain English. AI-powered web scraping with no code, no selectors, and no per-site setup.

Influencer Search β€” AI-Powered Discovery by Influship

influship/influencer-search

Search for Instagram influencers using natural language. Find creators by niche, location, audience size, engagement rate, and content style using AI-powered semantic search. Powered by Influship.

17

5.0