Launching today

Samiksa
Understand research, test it yourself, publish your own
29 followers
Understand research, test it yourself, publish your own
29 followers
Samiksa is an AI co-researcher. Give it a paper, blog post or any document, and it turns it into an interactive document shaped by your questions, not by how it was written. Ask what you want to know and the page reshapes around the answer, each claim traced to its source. Then launch a sandbox experiment: rerun what the research shared, or test your own hypothesis as your co-researcher helps with code and data. Your own work goes from rough idea to research you can publish. Join the waitlist.









Payment Required
Launch tags:Productivity
Launch Team / Built With

AWS Activate CreditsCredits that fuel your startup’s next milestone
Promoted



Hi Product Hunt, I am Mihir, the maker of Samiksa.
Anyone who does research knows the grind that comes with it. Reading a paper is the easy part. Finding out whether its result holds, and then getting from your own rough idea to something you can test and write up, is where the weeks go.
Samiksa brings research to life. A paper, blog or any document stops being a static PDF, website and becomes a dynamic, interactive document: you can question it, rerun it and change it, and an AI co-researcher takes on the mundane work.
How it works
Load any research. Give it a paper, a blog post or any document. Instead of reading it the way it was written, you explore it the way you want: ask a question and Samiksa reshapes the document around the answer, with every claim traced to the table, section and page it came from.
Explore it. The document responds to you. Turn a table into a chart, open the source behind a claim, or ask your co-researcher about any block.
Launch a sandbox experiment. When the research shares its data and code, run it to reproduce the original number, then change the setup and see your result beside it. Not sure what to try? Ask your co-researcher to change it for you.
Publish your own. Start from a rough idea. Samiksa turns it into a hypothesis you can test, runs the experiment, and draws the diagram from your code.
What to know before you try it:
Samiksa is in early access. You join a waitlist at samiksa.app; it is not open to everyone yet.
The video and images use a sample paper with recorded runs.
It does not replace your judgement. Nothing in your paper changes until you accept it.
I would like to hear from people who read or write research: which part of your week would you hand over first, and what would make you trust the result?
What does it do that dropping the PDF into Claude doesn't?
@thomas_wazza , it's a good question! I'd say Samiksa takes a different approach from Claude in three ways, because it's built around the researcher's experience:
When you drop a PDF into Claude chat, the paper goes into the conversation and answers come back as text. Samiksa flips that: the paper stays at the centre and the AI works around it. It reshapes the paper around your question, so you don't have to follow the order it was written in.
Idea is grounding is core aspect, every hypothesis, result and experiment links back to where it comes from in the paper/citation. In a chat, you have to ask for that and then go and check it yourself.
It is bit opinionated/tailored toward researcher. It already knows how researchers like their answers. Often you just want to understand a result quickly. Samiksa has tools built in for that: switch a table to a chart, rerun a quick experiment,(if there) and follow it back to the hypothesis and the original citation. In a chat, you'd have to ask for each of those every time.
To your trust question, rerunning the original number before changing anything is what would do it for me. A reproduced baseline is the difference between "the paper says" and "I checked."
What happens when a paper doesn't share its code or data? Does Samiksa say plainly that the result can't be reproduced, or does it try to rebuild the setup from the text?
@ssindhia Samiksa only reproduces a result when the code or data is actually there.
If the paper doesn't share them, Samiksa doesn't try to rebuild the setup from the text. In most of those cases you only have the method as written, so it explains the proposed approach and what the paper claims, tied back to where the paper says it. Because Samiksa builds its interface around each paper, it simply won't show a block for running or testing when there's nothing to run.
Building a test setup from the written method alone is something we might look at later. It brings its own problems, like whether you can trust code the AI wrote and who reviews it, so it's not something Samiksa does today.
letting the question reshape the document is interesting
the source links and the experiment are what would let me check the answer rather than just read a cleaner summary