Normain
Trusted insights from complex documents
943 followers
Trusted insights from complex documents
943 followers
Normain is an extraction-first AI for complex documents. It delivers structured, traceable insights grounded in source material - designed for validation and reuse, not chat-based summaries that hallucinate.
Free Options
Launch Team / Built With
Render: Get up to $100K in startup creditsCloud infrastructure for founders who want to move fast and NOT break things
Promoted
I've been trying Normain out for a couple of days now – incredible! How are you ensuring the AI is getting the right answer?
@elof_gerde1 Thanks for being an early supporter Elof! We have a state of the art agentic engine that goes above and beyond what's possible with just RAG. Normain uniqely interprets both the question and the input data, plans out an analysis path, and evaluates itself along the way. The result is remarkably accurate. Then, in addition, we have the Normain Trust Panel that puts human validation at your fingertips.
@elof_gerde1 Happy to hear this, and thanks for the question.
@sara_landfors You're spot on in the answer to one of the key differentiators that makes Normain so good. How we are able, through the Trust Panel and the advanced agentic technology, to ensure the AI is getting the right answer.
Congrats on the launch! I'm curious about how fast it is. Let's say I have 100 PDF purchase orders and 100 delivery confirmations and I want to cross-check which purchase orders have delivery confirmations that match precisely, and flag any discrepancies (either it was not possible to match the order with a deliver confirmation at all, or the content in the delivery confirmation does not match the purchase order details). How long time would that take, roughly? Could you make that kind of operation and response available in an API?
Totally understand if this is not a use case that fits what you're looking to build! But figured I'll ask 😄 All the best!
Hi@fredrik_olovsson ! Thank you, the Normain team has been working hard on making the launch happen. 👩💻
Your use case fits Normains capabilities well, especially the repeat mode which allows you to extract insights for each file you have uploaded.
My suggestion is to upload a single order with all your confirmations and confirm Normain is able to extract the correct confirmation for that order (or whatever else information you need to be certain the order has a confirmation). When you are satisfied with the insight you can repeat that extraction with as many orders as you want. 🔁
We do not have the extractions available as an api yet, it's on the roadmap though so we will keep you posted regarding that!
@fredrik_olovsson Hey Fredrik, excellent question! It's hard to estimate the time, but maybe 10 minutes max? I encourage you to try it out for yourself. We're working on functionality to be able to import "questions" from a spreadsheet, which I would assume is where your delivery confirmations sit? This is definitely a use case we want to support, so eager to learn more.
@fredrik_olovsson Thank you so much on the congrats. Let us know if you got your question answer 😊 We're here to answer all your question and act on feedback
@Normain great launch! congrats
how did you achieve precise extractions from large documents?
@ponikarovskii Heey thanks so much, and great question! It's a three step solution:
Purpose-built document interpreter that doesn't lose a single nuance, even from poor-quality scanned pdfs, tables and graphs
Agentic engine that builds its analysis path on the fly—making sure it doesn't stop before it's sure the answer is as good as it gets
Normain's Trust panel: makes it super easy for humans themselves to validate the outputs.
Curious to hear your feedback!
@ponikarovskii Thanks for a great question!
@sara_landfors
When the system detects conflicting data points across different file types, does it flag the discrepancy for manual review or does it prioritize one source over the other?
@valeriia_kuna Excellent question! Normain will flag this in the Trust Panel (shows up automatically when you click an insight). If there are conflicting data sources, Normain will put a lower confidence score (Medium or Low). We're working hard on improving the validation workflow even more, and would love to hear if you have more feedback?
See more details in 3. Extract and validate on our docs page here.
@valeriia_kuna your feedback will go straight to the product roadmap in order to make sure the Trust Panel stays at world class
@sara_landfors
@sara_landfors @dennislandfors
Happy to have contributed to the roadmap! Looking forward to see how the Trust Panel will develop!
Amazing, big congrats on the launch @sara_landfors @dennislandfors and hardworking team! Reliability is incredibly important to trust our AI tools especially as knowledge workers. Being in finance, one big obstacle to fully integrate tools has been hallucinations and not fully trusting the accuracy of the date and output. Looking forward to testing more advanced analysis!
@sara_landfors @shahrzad_moadeli Thank you for these words. Very kind of you 😇 It's spot on! I look forward to receiving your feedback after you've done some more testing on advanced analysis.
@sara_landfors @dennislandfors @shahrzad_moadeli This is amazing hear! Thank you Shahrzad!
Congrats! Seems powerful but how do you enforce zero hallucinations?
@daniele_packard Thanks for the great question Daniele. It's a three step process:
First, we parse input documents with a state-of-the-art degree of accuracy. We preserve all detail in tables and graphics.
Second, we have a truly agentic system that really understands your questions, and grounds answers only in input data.
Third, we have Normain's Trust Panel that puts the human expert from and center in validating the AI generated output in a user friendly and structured way
Would love to hear how Normain performs on your trickiest analyses!
@daniele_packard Thanks for a great question! 🎯
@sara_landfors
Been following Normain for a while, awesome to finally see this live here 🚀
Big kudos to Kalle, Dennis, and Sara, this clearly comes from real experience with years in AI building and iteration, not just something put together to ride the AI hype. Feels like a tool built for people who actually use this stuff day to day.
Love the focus on pulling out what actually matters from complex documents, without losing trust or context. The fact that insights are traceable back to the source is a huge deal and something most AI tools still get wrong.
Congrats on the launch! Excited to see where this goes 👏
@josef_karakoca Thanks so much for the shoutout! Really great to hear you see the value in the product and keep using it. Once you've seen the Trust Panel for validation, it's hard to go back to the old Conversational AI-mode. What are your best use cases?
Agree with @sara_landfors , thank you so much@josef_karakoca for this shoutout. It means a lot with your support!