PaperMachine

The AI analyst that shows its work.

5 followers

The AI analyst that shows its work.

5 followers

Visit website

Data visualization tools

•

AI Data Scientist

The AI analyst that shows its work. Runs Python and R on your own machine, and traces every chart back to the code behind it.

Free

Launch tags:Analytics•GitHub•Data & Analytics

Launch Team / Built With

OpenAI

GPT-6 Astra Challenge

How did Astra change the scope or ambition of what you built?

Maker

Astra shifted our ambition from building an AI that can answer analytical questions to one that can carry complex, multi-step analysis through to completion. That greater autonomy made provenance even more important: PaperMachine keeps every result connected to the code, logs, data, and environment behind it. Astra expands what the agent can do; PaperMachine makes that autonomy inspectable and accountable.

Report

23d ago

Maker

📌

Hey Product Hunt PaperMachine started from a question that kept bothering us: as AI becomes capable of doing more of the analysis itself, how do we know what actually happened between the question and the chart? Most AI tools optimize for the answer. We wanted to preserve the process. PaperMachine runs Python and R locally and turns an analysis into something you can inspect rather than simply trust. Every step can expose the code that ran, its output, and the state of the kernel. Every figure, table, or file carries provenance back to the code, execution log, conversation, and environment that produced it. If a human edits a chart, that becomes a new version rather than silently rewriting the original. That distinction matters to us. In research and serious data work, a plausible result is not enough. You need to be able to ask: Which data was used? What changed? What code ran? Under which environment? And can another person reconstruct the reasoning path? For the GPT-6 Astra Challenge, we are exploring the next step of that idea: pairing stronger long-horizon reasoning and multi-step execution with an analysis environment designed to keep the resulting work inspectable. We don't think the future of AI analysis is simply “more autonomous.” We think it has to become more autonomous without becoming less accountable. We'd especially value feedback from researchers, analysts, and builders: what would you need to see before you were willing to trust an AI-generated analysis enough to build real decisions or research on top of it? Thanks for taking a look at PaperMachine.

Report

28d ago

Forum Threads

p/github-502

•

23d ago

What would it take for you to trust an AI data analyst?

Hey everyone! We're launching PaperMachine on Product Hunt tomorrow (Sept 18) as part of the GPT-6 Astra Challenge.

The problem: AI data analysts are black boxes. You ask a question, you get a chart but when a reviewer asks "how did you get this number?", the answer lives nowhere. For researchers and analysts, that's a trust problem, not a convenience problem.

PaperMachine is a desktop app that runs Python and R on your own machine. The difference: every step shows the exact code it ran, and every chart traces back to the code, log, and environment that produced it. Nothing arrives out of a black box.

Built for this challenge with GPT-6 Astra itself a full desktop app with local kernels, artifact provenance, and chart editing, assembled through Astra's multi-step agentic coding. The app is its own demo.

View all