HACKER Q&A
📣 langtang1996

How should I improve my product?


I developed a saas software (https://www.chatpapers.ai), aim to improve reading experience, which has two main features: chat with pdfs & generate mind maps. The question is: Most of my users are interested in mind map generation but not chat with pdfs, many of them type commands in the input area (which is designed to ask questions about the pdf) and try to generate a mind map according to their commands, but this feature has not been developed. How should I improve it?

I have 3 options:

  1. Add intent detection to the chat module, support both mind map generation with commands & retrieve information from the pdf in a same input box, which may increase the latency and take wrong actions.

  2. Remove the chat with pdf feature and turn the product to mind map generation, which mignt strange because the product name is chatpapers.

  3. Design a new product to focus on mind map generation, and keep this product as it is.
Which could be better? Or any other suggestions?

  👤 JohnFen Accepted Answer ✓
This is just what I'd do -- not saying it's the best thing to do.

With a new product, you're still working out what aspects of it appeal to customers more than others.

My habit has been to find that out as quickly as possible, then focus the product to just do that as well as possible. If I were in your shoes, I'd focus like a laser on the mind maps. Once that's nailed down and polished, I'd consider bringing in the other aspects.

The advantage of doing this is twofold: by focusing on a single thing, you can really excel at that thing and attract more customers because it rocks so much. Then, once you have solid customer base, you have a receptive existing audience for expanded capabilities (and to guide you as to what form is best for those new capabilities).

👤 caprock
Why not ask the customers? It's great you have some traction and their actions are leading you to a direction where they find value. Lean into that!

In any event, no need to make drastic changes. Maybe idea 1, but definitely not wholesale changes like 2 and 3. There are plenty more options on the table, such as:

* make two submit buttons in chat area (chat vs mindmap)

* support very specific mindmap commands and use simple string matching for intent detection (not models)

* create a way for people to choose either chat or mindmap commands for the right hand sidebar

* add more example inputs (like the "show me a summary") to guide the users

* etc

👤 nonrandomstring
I live by the Unix philosophy, so one simple tool that does one thing very well is how I roll.

I could literally see myself on the command line typing

pdf2mm -i report.pdf -o report.xml -f "sales, drivers"

where -f is a list of strings to direct focus (seed the "centre of the map")

Dump .xml or .opml

Get this right and you'd have no problem offering that as a paid Web service and throw the cmdline util out there for free please. :)