3 ms·

Show HN: Nexa, AI-powered analytics for startups

6 points by zahmed1 3y ago

Excited to share Nexa's public beta and get feedback from the HN community! -- The Problem Companies today spend way more time querying and analyzing data than acting on it. That needs to be flipped. At startups without data teams (who we’re built for), engineers waste precious time writing SQL queries for others (e.g. Growth, Product, Sales), and then those teams spend hours manually analyzing and formatting the data to make decisions. All of this is time we can give back with AI. -- Introducing Nexa With Nexa, connect your MySQL / PostgreSQL database or upload CSVs and extract insights in plain English, not code. These insights can be added to Notion-page like dashboards and shared with your stakeholders instantly. Behind the scenes, our AI SQL agent writes queries, generates summaries, creates visualizations, and suggests prompts to help you uncover insights you didn’t think of. You might also find that uploading a CSV and exploring your data in Nexa is much faster and more pleasant than writing IF statements in sheets and copy / pasting charts and tables from sheets to docs ;-). -- Try it out and let us know what you think. And a HUGE thanks for the support and feedback!

jaggs 3y ago
This actually looks nice, and very useful. I'm not sure I need to see the SQL every prompt, maybe a toggle to switch it on and off? But looking forward to playing some more.
- zahmed1 3y ago
  Thanks for trying it out and for the feedback! Good point, that's something we can easily hide and handle in different ways (e.g. show as a suggested follow-up prompt, toggle on / off, etc.)
XCSme 3y ago
The example with "interesting insight" works well with a small dataset (CSV) that fits into the AI context memory, but how do you deal with a large dataset (e.g. a database of tens of GBs), the AI can't read/process all that data? Does it just do some "random" queries to generate some facts?
- zahmed1 3y ago
  Hey, yes, this gets more complicated as datasets get larger. Our early customers have simple datasets (using spreadsheets, csvs, and relatively small postgres or mysql dbs), so it works very well for them. As we continue to move to larger customers, data warehouses, and more complex use cases, we'll build semantic layers (aka. provide a lot of context on the db, tables, business metrics, sample queries, and more), use RAG alongside our SQL agent, introduce more robust auditing / RLHF, etc. All solvable as our customer needs require it. Also, context window limits continue to increase, and since we're plugging into the best models to solve customer problems (not building our own), tbd on how much of what I've listed above (e.g. RAG) we'll need to do when the time comes.

6 points by zahmed1 3y ago

jaggs 3y ago

This actually looks nice, and very useful. I'm not sure I need to see the SQL every prompt, maybe a toggle to switch it on and off? But looking forward to playing some more.

zahmed1 3y ago
Thanks for trying it out and for the feedback! Good point, that's something we can easily hide and handle in different ways (e.g. show as a suggested follow-up prompt, toggle on / off, etc.)

XCSme 3y ago

The example with "interesting insight" works well with a small dataset (CSV) that fits into the AI context memory, but how do you deal with a large dataset (e.g. a database of tens of GBs), the AI can't read/process all that data? Does it just do some "random" queries to generate some facts?

zahmed1 3y ago
Hey, yes, this gets more complicated as datasets get larger. Our early customers have simple datasets (using spreadsheets, csvs, and relatively small postgres or mysql dbs), so it works very well for them. As we continue to move to larger customers, data warehouses, and more complex use cases, we'll build semantic layers (aka. provide a lot of context on the db, tables, business metrics, sample queries, and more), use RAG alongside our SQL agent, introduce more robust auditing / RLHF, etc. All solvable as our customer needs require it. Also, context window limits continue to increase, and since we're plugging into the best models to solve customer problems (not building our own), tbd on how much of what I've listed above (e.g. RAG) we'll need to do when the time comes.