5/19/2025 at 4:14:26 AM
Hey HN,I’m one of the developers behind https://capalyze.ai, an AI-powered tool that helps small teams analyze their data using natural language — no code, no templates.
The idea came from conversations with indie e-commerce sellers and creators who had _tons_ of data (spreadsheets, reviews, exports from marketplaces) but lacked the time or tooling to make sense of it.
With Capalyze, you can:
1. Ask questions like _"What are the most common complaints in these reviews?"
2. Paste or upload product data and get a summary of pricing trends
3. Generate charts, compare columns, or extract keywords — just by asking
It works best with your own datasets right now (CSV, Excel, etc.). Web scraping isn’t live yet— we’re actively building it, and you can follow our updates if that’s important to you.
We’ve tested it with early users in e-commerce, real estate, and content — and the feedback has been super helpful. One user called it “ChatGPT with a purpose.”
We’d really appreciate feedback from the HN community:
1. Is the interface intuitive?
2. Are the responses helpful and explainable?
3. What would make this more useful for you?
Here’s the link: https://capalyze.ai Happy to answer questions and chat more about how we built it (multi-model backend with OpenAI, Claude, DeepSeek, plus a simple orchestration layer).
Thanks!
by alexliu518
5/21/2025 at 12:14:53 PM
My initial thought is "this can't possibly work."We don't even have text to SQL working properly, and excel is so much messier than that.
What simplifying assumptions are you making about the spreadsheets people send you? How do you ensure correct results?
by pbronez
5/22/2025 at 8:49:53 AM
Very critical question. Excel is indeed more complex. Before analysis, Capalyze first preprocesses the data, which is crucial. We have designed a set of preprocessing algorithms that essentially focus on how to better identify the data suitable for analysis in Excel and clean and repair it. This process also leverages LLMs, as we found that LLMs perform quite well in recognizing table structures.by alexliu518
5/22/2025 at 8:09:49 AM
How does the chart generation work under the hood? It's quite magical. Also how did you build the spreadsheet interface it's very cool.by helltone
5/22/2025 at 8:50:19 AM
Two main aspects: 1. How to handle the data related to the target problem; 2. Choosing suitable charts to present this data. #1. By leveraging the increasingly powerful coding capabilities of LLMs, we can appropriately process raw data to obtain a dataset that closely aligns with our goals; #2. We expanded echart and utilized its rich chart types already supported, along with the Univer SDK from the Univer team, ultimately creating tables.by alexliu518