@david_morse thank you! we optimized creating value for folks who need it the most! we're very thankful for our early customers.

@benjamin_beltzer exactly! that's a very good use case. we're talking to partners in the industry to learn needs/pain points.

Hey @eriktorenberg — happy to provide background. From the beginning we’ve had a thesis that accelerating the advancement of AI required expanding beyond our initial AV market. As we talked with a variety of customers, we quickly realized that the document processing market was very fragmented w/ many pain points, primarily quality, cost, and elasticity. Off-the-shelf OCR engines don’t offer the high-quality desired, and often require complex biz rules to be added on top (for example, linking an address to a specific name mentioned). If further quality verification is needed, customers often recruit their own people to verify the results of an OCR model, which requires massive eng and ops investment. Even if customers are able to set this up, they’re often vulnerable to peaks and valleys in data needs, which results in unfulfilled demand or sunk cost. Building off of our work in AVs, an industry that demands very high data quality, it was natural to apply the complex ML + human review workflows to this problem. Using similar techniques applied to image, video, and LIDAR labeling, we can mitigate bias that comes from ML results while reducing unnecessary costly review (for example, by skipping human review on an annotation that had high confidence from the ML model). Combined w/ our massive WFH workforce, we’re able to offer customers a seamless experience w/o having to make the difficult trade-off b/t quality, cost, and elasticity.

@jlff thank you! Feel free to put us in touch with them! melisa[at]scale

Hey @chaitanyadesai, great question -- as mentioned in an another reply, off-the-shelf OCR engines don’t offer the high-quality desired, and often require complex biz rules to be added on top (for example, linking an address to a specific name mentioned). If further quality verification is needed, you can recruit your own people to verify the results of an OCR model, but it requires massive eng and ops investment. Even if you're are able to set this up, you'll be vulnerable to peaks and valleys in data needs, which results in unfulfilled demand or sunk cost. Building off of our work in AVs, an industry that demands very high data quality, it was natural to apply the complex ML + human review workflows to this problem. Using similar techniques applied to image, video, and LIDAR labeling, we can mitigate bias that comes from ML results while reducing unnecessary costly review (for example, by skipping human review on an annotation that had high confidence from the ML model). Combined w/ our massive WFH workforce, we’re able to offer customers a seamless experience w/o having to make the difficult trade-off b/t quality, cost, and elasticity.

@duellsy Thank you, Chris! Let us know if you have any feedback. We're always looking forward to talking to folks who have worked on in similar area.

@duellsy Thanks Chris! If you're still in the market for these products, feel free to check it out for even more partnerships between our companies 😀