caoxhua

Thanks! Some competitor did pretty well, the total addressable market should be huge enough, for example Intuit Mint claims to have 10M (free) users, some funded companies benefits a lot from its shutdown.

yeah, opportunity cost is huge.

There are some big parts identified for this case, and because it is dependency of 3rd party service, we can not do much about it.

But more importantly, I think it is the product itself, the market is probably does not appreciate that much of "customization / personalization" as I thought, in other words, we are probably haven't figured out the best path to get the ideal users yet.

In the other side, free users do not have this issue. We do have free tier which counts for 1/3 of all active users.

How do you tell product market fit? we have about ~1K paid users, people constantly send positive praises, but also many people churned away.

I doubt if "bootstrap" is the right way for this type of product/market.

The famous PMF survey (how do you feel if you can no longer use this), %50.8 responds with "very disappointed". And I deeply know this product is pretty unique offering in "finance tracking" market.

But the MRR stays in this level for 6 months now. I feel I do not have ideas to break it through.

Excel/Spreadsheet are best invented software ever.

However, the other side of this double-edge sword is its fragility, when data or logic grow in complexity, it raises more errors than it should. That's when a niche product gets its chance to chime in.

Have you heard of Fina Money? SpreadSheet works, but Fina tries to make it easier to analyze and visualize your finance data from sheets + live banks. It solves the problems you mentioned.

I think Fina Money is the most flexible one, normally a good fit for engineer mindset.

genius, haha~

I always click Y-axis to modify the axis range, usually modify low-value to 0 and it looks better - https://columns.ai/chatgpt/uUN2oYp7Yusstk

hahaha~

hahaha~

Trusting is a problem, to be honest, mostly for fun. If you want to trustable source, feel free to copy text paragraph and let it parse for you, there is another mode for that.

Yeah, not mobile friendly UI. because the visualizer really needs some space to render, so overall I didn't pursue to optimize it for mobile.

Anything is clickable and styleable on the canvas though.

So it's a pretty simple wrapper of LLM model in use (currently gpt-4o), it does not add much technical stuff in it.

It does not use database for any "random search", but yes, columns.ai is a data analytics tool that allows you to connect supported live data sources like Google Spreadsheet, Airtable, Notion Database to create visual stories.

The analytics engine is home built (https://github.com/varchar-io/nebula) but it is not a database. And I don't use LLM agents, just build logic how to purify data returned by LLM, and fit them into an optimized visualization.

Hope I answered your question!

If I click the `Y-axis`, then set the low-range to 0, then it looks better https://columns.ai/chatgpt/WoGchDeDpKNKWl

this is what I saw, https://columns.ai/chatgpt/jMTu9QXU9nbeCs "no fact check", not sure if LLM would give source cite while giving those numbers.

Unless it's very "public common", otherwise, the numbers can be used for trustworthy work. In general, I think this is just for "fun play" most of the time.

I don't why it's flagged, it's just a free service for people to play with.

why not? it is necessary to protect API server with secure auth. It does not charge you a penny.

We have open sourced the core data model and graph model used in Columns. Now this simple API is ready to play with, check it out now.

"I do it in Python outside GA." - do you mean you export different segments data from GA and they use Python to glue them for combined/join view?

Fundamentally they both need to do computation to answer query, but they are very different, a couple of highlights: 1. API service invest a lot on reliable data ingestion/loading for users, you specify data spec, data will be served with lease term reliably. It takes care of all the data sync maintenance for you. 2. It works seamlessly with real-time streaming such as Kafka for RTA. PG is not suitable for this. 3. It's distributed nature + In-memory store guarantees low-latency API service.

Hope this describes a reasonable overview to think of it.

we use our in-house baked engine - open sourced here https://github.com/varchar-io/nebula

Yeah, Tinybird has lots of similarities, I will do more research on it, thanks for the reference.

Thank you for the correction, appreciate it! I'll update it.

Hey, thanks for your comment!

You don't encrypt, just leave your data where it is, what we need is a live connection, after loading your data in our distributed cluster in cloud, we have option to encrypt the data for further security if needed. So yes, we encrypt at our end if asked.

Our API service doesn't store data, only load them into distributed memory for computing and API serving.

Let me know if you're interested in this API service, would like to see if we can help.

You already said "on the columns", so it will be https://columns.ai, :)

cool - I also hand crafted a CSV parser following RFC4180 a while ago, not sure if you have a repeatable way to benchmark the performance difference?

https://github.com/varchar-io/nebula/blob/master/src/storage...

very cool!

Somehow it makes me think of "Tinder for topic" - focus on a mobile app which uses ML algo to keep learning users' "keep" or "throw-away" to keep improving users' feed, I believe it have potential for a big hit.

It interested, I'm open for more brainstorm, :) (email: cao@columns.ai)

Thank you for your feedback, appreciate it! yes - for simple, we can call it Pinterest for data stories.

I haven't thought about that problem of unclean data yet, right now as prototype, the map module drops the unmatched entry after trying to identifying it as state (full or short names), county names, zip code, etc. In the future, editing distance might help improve matching.