Metatext AI - Classify and Extract Text Reviews

Metatext has greatly fit my business need by efficiently classifying and extracting information from text and documents. I am impressed with the accuracy and reliability of your customized language models, which are tailored to my specific needs. It has greatly improved our workflows and saved us valuable time and resources. Please keep your good work!

I found it difficult to use Metattext initially. After I played around with it for a bit I have now fully stacked with 4 codes.

Why?
- In-app support is excellent. I must have asked a dozen or so questions and they were all answered quickly and competently
- the models do just work. it's super convenient to access these models without having to set up python code yourself. you won't need a coder once you figure out how things work
- features are being added quickly: text extraction is now live, more is coming
- documentation is now available: they now added a few videos and some instruction documents that provide much better explanations than previously

My sense is this: this product is solid. It's almos the opposite to some of the other products on appsumo: with Metatext the actual product is solid and works. The guys know what they're doing. They're PhD-level nerds who know their stuff.

Where they could get some help is with 'making it all look better': get a copywriter, hire someone to write some instruction manuals, that kind of stuff. This isn't hard though and I'm sure they will improve this over time.

Glad I'm fully stacked and looking forward to more additions.

This is enabling a few services to generate or analyse content for multiple scenarios. The sentiment analysis alone makes this a valuable investment if you want to assess the potential of sales, investment, idea definitions. Really impressed so far.

This is a pretty unique product. Just full stacked.

As a non-dev, I never really expected to be able to train my own model without a lot of headache. If I want good data AI capabilities, I spend most of my time iterating prompts and I can only get it so far. I uploaded 150 leads (that I manually reviewed / categorized) and it was just as good as GPT 3.5 in categorizing my leads. Once I get to a 1,000+ I bet it's going to be stellar.

Really excited to see where this product goes. A couple comments / questions to the team:

- Would be great if I could indicate multiple data columns to train on. This caught me off guard with the onboarding (kept trying to select more columns). Sure, I can concatenate the columns into a single column but why? And wouldn't data accuracy be better if you trained on multiple distinct columns? This was the part that felt the most off to me.
- Also, I am using the same data set for predicting multiple categories (ie multiple data columns). Could this all be lumped together in one model eventually or do I need to upload the same data for each categorization? There seems like some potential efficiency here for your users and could avoid a big headache as well (especially when the data set gets big).
- I think your existing NLPs could be a hidden gem if you get some really on point ones. The list didn't really jump out to me but I can see this being big as your list grows
- Onboarding guide could use a little smoothing over.

And to that 1 taco review, wtf?