And the fact that we’ve built a 100 amp-hour stacked pouch cells doesn’t mean that we will end up actually putting 100 amp-hour cells inside the cars, it’s most you understand the boundary of the limitations. The final cell might very well be a 70 amp-hour cell. But all this we’ve built A-samples, and also B-samples, and then we improve the quality, and then increase the production output so that we can collect more data and then also run more experiments faster so we can make all these — so we and our OEM JDA partners can make all these decisions based on data. So, we haven’t — and that’s what we’ve been focusing now, stacked pouch cells, maybe stacked prismatic cells, and then also with different cathodes, and also just the capacity of the cells, we’ll also test a range of different capacities so we can find the optimal final cell design that we can actually put into the vehicle based on a complete — a very holistic consideration of lots of different factors.
Gabe Daoud: Got it, okay, that’s helpful. And then you mentioned data, so maybe a good segue into a question on Avatar. You know that some of the challenges that actually enabled you to drive an improvement in tracking or predicting cell safety on the larger amp hour cells with 60% accuracy versus 0% in the beginning of the year, so — and then the smaller cells have 99% accuracy. So, could you maybe just talk a little bit more about Avatar and the progression there and some of the improvements there that you’re seeing?
Qichao Hu: Yes, so we really found Avatar is a necessity. And then beginning of last year, beginning of 2022, because we didn’t really have any large cell data, and Avatar is basically very data driven. So, we had zero capability to predict anything to the larger cells because we just didn’t have any data to begin with. And then we collected more data, and then these larger cells are actually quite different from the smaller cells. For example, the impedance is much lower, and then any variation in impedance — you know, lot of the parameters that we use to predict health and life in small cells just simply don’t work when we scale up to the larger cells. So, we had to develop new parameters. And then also, we also had to collect a wider spectrum of data, for example, the smaller cells we did current, voltage, and temperature.
Now, we’re doing current, voltage, temperature, and pressure. Pressure is a really important factor in predicting the health and safety of these large lithium-metal cells. And then also based on the data that we collect, we also both through a combination of internal physics-based models where we actually predict health based on actual fundamental physics mechanism as well as just machine learning models, that’s really recommended; very interesting parameters. So now, with more data, so last year, 2022, we had B-sample lines — A-sample lines, and then we built and then tested about a thousand of these cells now. And then based on that data pool we got to cross 60% accuracy. Now that as we prepare to enter B-sample with much better quality, much better consistency, and much higher output, that we expect both the quality of the data as well as the size of the data pool to significantly grow.
And then that will significantly improve the accuracy of Avatar for the larger cells even faster.
Gabe Daoud: Got it. Thanks, Qichao, thanks everyone.
Qichao Hu: Thanks, Gabe.
Operator: Thank you. Our next question comes from Shawn Severson from Water Tower. Shawn, your line is now open, please go ahead.
