Promptrepo
Build your fine-tuned AI model using data in Google Sheets
148 followers
Build your fine-tuned AI model using data in Google Sheets
148 followers
Build your own AI model for classification, extraction and generation—right from Google Sheets or the no-code Ai model builder. Test it using the intuitive forms UI or integrate it in your apps, websites & workflows using the API and launch AI features quickly
Guesswork
One week before my Y Combinator interview, my doctor told me I needed an open-heart surgery. I found myself explaining my condition again and again to different doctors across multiple hospitals. Filling out forms and getting appointments became a full-time job. I remember thinking: what if I could just take photos of all my reports, email them, and the hospital system could auto-extract everything? That’s exactly what we’ve built.
With Semantic Email, a patient can take a photo of their insurance card or referral letter, send it via email, and AI automatically fills out a Google Form, syncing data to Sheets and Drive — ready for integration into systems like EHR. While healthcare is personal to me, this works for any business on Google Workspace. Customers prefer email. Businesses need structured data. Semantic Email bridges that gap. (More here)
Would love to hear where you think this could help outside healthcare.
Tidyread
Impressive idea, Mani! 👏 The integration with Google Workspace is a smart move. Does the AI require any specific training to work with different types of documents, or is it ready to use out of the box? Looking forward to seeing how this evolves!
Guesswork
@Thanks! It works out of the box for attachments like pdfs and images. No training needed. But users can also ask questions about products or services via email. In that case, adding them as quick links helps the AI to answer better.
Smoopit
@manidoraisamy Really clever approach, using email as the interface removes so much friction. The Google Workspace integration makes total sense for adoption. I'm curious, what's the accuracy rate like? And how do you handle edge cases where the AI can't extract everything cleanly?
Guesswork
@rachitmagon Thanks! We use Gemini with MCP, which gives us native support for photos and PDFs. The accuracy is 95%+ for documents with clear text. For edge cases where the AI can’t confidently extract everything, we do two things:
Review and submit: We email back a summary with whatever was extracted. The user can fill the rest and submit.
Follow-up email: We follow up with the sender for mandatory missing info, conversationally, based on your form logic.
So the workflow always completes, even when the AI isn’t 100% confident.
Smoopit
@manidoraisamy Exactly, the fallback mechanisms are crucial. No point in having high accuracy if the system breaks down on edge cases.