[
  {
    "startMs": 1309,
    "text": "[music]"
  },
  {
    "startMs": 12720,
    "text": ">> My name is Bo. I'm going to be here"
  },
  {
    "startMs": 14800,
    "text": "presenting real-time voice agents with"
  },
  {
    "startMs": 17600,
    "text": "Frontier Intelligence. Effectively,"
  },
  {
    "startMs": 19760,
    "text": "going to be talking a little bit about"
  },
  {
    "startMs": 21600,
    "text": "how we at Xnor.ai architected our voice"
  },
  {
    "startMs": 24600,
    "text": "agent harness to get real-time voice"
  },
  {
    "startMs": 27880,
    "text": "with the Frontier level of intelligence"
  },
  {
    "startMs": 30400,
    "text": "that we need."
  },
  {
    "startMs": 32960,
    "text": "Okay. So, before I start,"
  },
  {
    "startMs": 35480,
    "text": "I think I wanted to kind of draw some"
  },
  {
    "startMs": 37440,
    "text": "parallels about"
  },
  {
    "startMs": 39200,
    "text": "why we decided to go with cascaded voice"
  },
  {
    "startMs": 42040,
    "text": "agents and especially kind of"
  },
  {
    "startMs": 45120,
    "text": "comparing that to self-driving cars"
  },
  {
    "startMs": 47040,
    "text": "which I was working in before. So, to"
  },
  {
    "startMs": 50000,
    "text": "me, cascaded voice agents makes a lot of"
  },
  {
    "startMs": 52720,
    "text": "sense when you view it in lens of kind"
  },
  {
    "startMs": 54520,
    "text": "of breaking it down into perception"
  },
  {
    "startMs": 56680,
    "text": "which is"
  },
  {
    "startMs": 58480,
    "text": "for self-driving cars, it's you know,"
  },
  {
    "startMs": 60480,
    "text": "the bounding boxes, the camera, the"
  },
  {
    "startMs": 62480,
    "text": "lidar."
  },
  {
    "startMs": 63720,
    "text": "For voice, it's going to be the"
  },
  {
    "startMs": 65600,
    "text": "transcription. Basically, effectively"
  },
  {
    "startMs": 67520,
    "text": "turning these like signals from the real"
  },
  {
    "startMs": 69080,
    "text": "world into"
  },
  {
    "startMs": 71120,
    "text": "elements of data that the language model"
  },
  {
    "startMs": 74120,
    "text": "or whatever brain you're working on"
  },
  {
    "startMs": 76640,
    "text": "can process."
  },
  {
    "startMs": 79200,
    "text": "Second one is the planning step which is"
  },
  {
    "startMs": 81600,
    "text": "pretty straightforward."
  },
  {
    "startMs": 83360,
    "text": "This is where the language model"
  },
  {
    "startMs": 85320,
    "text": "will take in the outputs from the"
  },
  {
    "startMs": 87440,
    "text": "perception stage and produce the outputs"
  },
  {
    "startMs": 90280,
    "text": "that you want to produce out back and"
  },
  {
    "startMs": 92840,
    "text": "out into the real world."
  },
  {
    "startMs": 94480,
    "text": "And finally, there's the controls layer"
  },
  {
    "startMs": 96960,
    "text": "where"
  },
  {
    "startMs": 98120,
    "text": "on self-driving, you'd be taking the"
  },
  {
    "startMs": 99520,
    "text": "trajectory that the planner would output"
  },
  {
    "startMs": 102400,
    "text": "and kind of turn it into the real"
  },
  {
    "startMs": 105360,
    "text": "controls to kind of build like drive the"
  },
  {
    "startMs": 107840,
    "text": "car. Here, we're turning the text into"
  },
  {
    "startMs": 111040,
    "text": "audio that we use to express our voice"
  },
  {
    "startMs": 113880,
    "text": "agent's thoughts."
  },
  {
    "startMs": 116760,
    "text": "And um"
  },
  {
    "startMs": 118440,
    "text": "yeah, so here I'm going to be like going"
  },
  {
    "startMs": 119840,
    "text": "to diving into each one of these"
  },
  {
    "startMs": 121520,
    "text": "elements and we've made a few kind of"
  },
  {
    "startMs": 124440,
    "text": "interesting tricks on each of these"
  },
  {
    "startMs": 126440,
    "text": "areas to"
  },
  {
    "startMs": 128119,
    "text": "improve the speed of our voice agents"
  },
  {
    "startMs": 130320,
    "text": "without sacrificing the intelligence."
  },
  {
    "startMs": 133520,
    "text": "So, the first one is going to be uh the"
  },
  {
    "startMs": 136280,
    "text": "transcriber layer. So, we came up with"
  },
  {
    "startMs": 138360,
    "text": "this concept called like the streaming"
  },
  {
    "startMs": 139920,
    "text": "speculative transcriber where"
  },
  {
    "startMs": 141760,
    "text": "effectively we are layering a fast"
  },
  {
    "startMs": 144360,
    "text": "streaming transcriber like Flux on top"
  },
  {
    "startMs": 147120,
    "text": "of or kind of below a"
  },
  {
    "startMs": 149520,
    "text": "uh scribe V2 or a accurate batch"
  },
  {
    "startMs": 151959,
    "text": "transcription which kind of takes in"
  },
  {
    "startMs": 154200,
    "text": "more context. It's a little bit slower,"
  },
  {
    "startMs": 155880,
    "text": "but it will give you more accurate"
  },
  {
    "startMs": 157600,
    "text": "detections."
  },
  {
    "startMs": 159080,
    "text": "So, we're going to walk through a"
  },
  {
    "startMs": 160640,
    "text": "scenario. So, in this in this case the"
  },
  {
    "startMs": 162760,
    "text": "agent just asked, you know, providing"
  },
  {
    "startMs": 164920,
    "text": "can you provide your name and date of"
  },
  {
    "startMs": 166440,
    "text": "birth and the user is going to say this"
  },
  {
    "startMs": 168840,
    "text": "and we'll see how that plays out um"
  },
  {
    "startMs": 171160,
    "text": "timing-wise. So, first we're going to"
  },
  {
    "startMs": 173040,
    "text": "get, you know, the short detection. Um"
  },
  {
    "startMs": 176640,
    "text": "we'll get it from we'll get it from the"
  },
  {
    "startMs": 177880,
    "text": "streaming layer. The accurate"
  },
  {
    "startMs": 180480,
    "text": "layer uh the corrective layer is not"
  },
  {
    "startMs": 182480,
    "text": "going to fire because it's the same"
  },
  {
    "startMs": 183760,
    "text": "text."
  },
  {
    "startMs": 184920,
    "text": "Um we're going to get some more"
  },
  {
    "startMs": 187360,
    "text": "streaming text detections and in this"
  },
  {
    "startMs": 189840,
    "text": "case the corrective layer is actually"
  },
  {
    "startMs": 191880,
    "text": "canceled because we got new um new text."
  },
  {
    "startMs": 194959,
    "text": "So, you know, more context, more audio"
  },
  {
    "startMs": 198320,
    "text": "is going to beat the old accurate one."
  },
  {
    "startMs": 201480,
    "text": "And here's where kind of the first"
  },
  {
    "startMs": 202959,
    "text": "correction comes in. So, because the"
  },
  {
    "startMs": 205400,
    "text": "scribe V2 layer understands, you know,"
  },
  {
    "startMs": 208200,
    "text": "the the context of the question, it's"
  },
  {
    "startMs": 210080,
    "text": "able to understand that this is talking"
  },
  {
    "startMs": 211440,
    "text": "about name and this is a date of birth."
  },
  {
    "startMs": 214320,
    "text": "Then a couple more detections, these are"
  },
  {
    "startMs": 215640,
    "text": "just punctuation, we don't care."
  },
  {
    "startMs": 217800,
    "text": "And so, in the end we kind of release"
  },
  {
    "startMs": 219920,
    "text": "this text over to the agent."
  },
  {
    "startMs": 224360,
    "text": "And moving on um to the language model"
  },
  {
    "startMs": 226760,
    "text": "layer."
  },
  {
    "startMs": 227800,
    "text": "So, here since we're kind of using these"
  },
  {
    "startMs": 230800,
    "text": "slow but intelligent LLMs, we really"
  },
  {
    "startMs": 233400,
    "text": "want to reduce the number of round trips"
  },
  {
    "startMs": 235240,
    "text": "and the thing that causes us to do a lot"
  },
  {
    "startMs": 238400,
    "text": "of inferences is tool calling. So, one"
  },
  {
    "startMs": 240960,
    "text": "way to get rid of that is by having"
  },
  {
    "startMs": 243520,
    "text": "background agents do the tool calling"
  },
  {
    "startMs": 246000,
    "text": "for you and kind of"
  },
  {
    "startMs": 248120,
    "text": "um push the tools back into the context"
  },
  {
    "startMs": 251040,
    "text": "of the main agent so that it thinks it"
  },
  {
    "startMs": 253200,
    "text": "made the tool call, but"
  },
  {
    "startMs": 255400,
    "text": "um"
  },
  {
    "startMs": 256000,
    "text": "but it it it really didn't."
  },
  {
    "startMs": 258000,
    "text": "So,"
  },
  {
    "startMs": 259320,
    "text": "uh so, we remember from like detections"
  },
  {
    "startMs": 261239,
    "text": "from before."
  },
  {
    "startMs": 262520,
    "text": "So, well, what happened is each one of"
  },
  {
    "startMs": 264720,
    "text": "these detections is going to trigger a"
  },
  {
    "startMs": 267880,
    "text": "um an early"
  },
  {
    "startMs": 270000,
    "text": "kind of generation of the agent and we"
  },
  {
    "startMs": 273160,
    "text": "but we won't actually"
  },
  {
    "startMs": 275200,
    "text": "emit this out until we're confirming"
  },
  {
    "startMs": 278160,
    "text": "that the user has finished speaking. So,"
  },
  {
    "startMs": 280360,
    "text": "in this case, the user says, \"Sure.\" The"
  },
  {
    "startMs": 282520,
    "text": "agent kind of knows that the user is"
  },
  {
    "startMs": 283800,
    "text": "about to say something else. Our"
  },
  {
    "startMs": 285400,
    "text": "background tool calling here, which is"
  },
  {
    "startMs": 287240,
    "text": "going to be helping us find figure out"
  },
  {
    "startMs": 289160,
    "text": "the name and the date of birth from the"
  },
  {
    "startMs": 290560,
    "text": "user detection, is not firing. So,"
  },
  {
    "startMs": 293880,
    "text": "nothing much there."
  },
  {
    "startMs": 295360,
    "text": "Um the next instant detection comes in."
  },
  {
    "startMs": 298120,
    "text": "It says that,"
  },
  {
    "startMs": 299640,
    "text": "you know, still not really a name. Um"
  },
  {
    "startMs": 302480,
    "text": "our agent kind of plays along and"
  },
  {
    "startMs": 303960,
    "text": "continues there."
  },
  {
    "startMs": 306240,
    "text": "Now, kind of a more more context come"
  },
  {
    "startMs": 308320,
    "text": "comes back. The agent kind of feels like"
  },
  {
    "startMs": 310400,
    "text": "there should be a name. It's going to"
  },
  {
    "startMs": 312000,
    "text": "ask to spell it out because it's"
  },
  {
    "startMs": 314080,
    "text": "probably thinking there's some"
  },
  {
    "startMs": 315320,
    "text": "transcription error here. Still no name"
  },
  {
    "startMs": 317800,
    "text": "or date of birth."
  },
  {
    "startMs": 319120,
    "text": "And then finally, this you remember this"
  },
  {
    "startMs": 320440,
    "text": "is kind of our corrected um final"
  },
  {
    "startMs": 322880,
    "text": "instant detection from the transcriber"
  },
  {
    "startMs": 325400,
    "text": "from the Scribe V2."
  },
  {
    "startMs": 327000,
    "text": "Um here, our eager kind of agent"
  },
  {
    "startMs": 330880,
    "text": "generation that was made without any"
  },
  {
    "startMs": 332520,
    "text": "tool calls is going to get canceled"
  },
  {
    "startMs": 334120,
    "text": "because the background agent finally is"
  },
  {
    "startMs": 335680,
    "text": "able to find the name and date of birth"
  },
  {
    "startMs": 337880,
    "text": "it's looking for. So, it's going to"
  },
  {
    "startMs": 340080,
    "text": "retrigger and now the the agent actually"
  },
  {
    "startMs": 342520,
    "text": "has the context it needs."
  },
  {
    "startMs": 344800,
    "text": "Um and you see here, it's kind of we're"
  },
  {
    "startMs": 346360,
    "text": "doing it the tool call here is a little"
  },
  {
    "startMs": 348000,
    "text": "bit um um"
  },
  {
    "startMs": 349760,
    "text": "some intelligence there. We're going to"
  },
  {
    "startMs": 351160,
    "text": "be like, you know, correcting"
  },
  {
    "startMs": 352840,
    "text": "mis-transcriptions of name, and doing"
  },
  {
    "startMs": 354840,
    "text": "some like phonetic matching here."
  },
  {
    "startMs": 357520,
    "text": "Um and yeah, and then we'll kind of"
  },
  {
    "startMs": 359840,
    "text": "once we've understood that this is the"
  },
  {
    "startMs": 362000,
    "text": "end of the user utterance, we'll kind of"
  },
  {
    "startMs": 363280,
    "text": "emit it out. So, pretty standard."
  },
  {
    "startMs": 366440,
    "text": "Okay, and then the next layer here is"
  },
  {
    "startMs": 368360,
    "text": "going to be text-to-speech. So, with"
  },
  {
    "startMs": 370520,
    "text": "text-to-speech"
  },
  {
    "startMs": 371960,
    "text": "the goal is to kind of take what the"
  },
  {
    "startMs": 373360,
    "text": "agent said, and the agent's going to be"
  },
  {
    "startMs": 375000,
    "text": "emitting this in a streaming fashion."
  },
  {
    "startMs": 377000,
    "text": "So, we're going to need to"
  },
  {
    "startMs": 379520,
    "text": "um produce audio as quickly as possible."
  },
  {
    "startMs": 381840,
    "text": "And ideally, what you can do is before"
  },
  {
    "startMs": 384680,
    "text": "the agent has even finished generating"
  },
  {
    "startMs": 386360,
    "text": "the full text you can have the audio"
  },
  {
    "startMs": 390200,
    "text": "play, so it's kind of hiding the latency"
  },
  {
    "startMs": 392280,
    "text": "of finishing the generation."
  },
  {
    "startMs": 394760,
    "text": "So,"
  },
  {
    "startMs": 396000,
    "text": "um I'm going to kind of play the"
  },
  {
    "startMs": 397120,
    "text": "streaming um"
  },
  {
    "startMs": 398840,
    "text": "the stream the streaming uh agent output"
  },
  {
    "startMs": 400880,
    "text": "now. So, starts with you."
  },
  {
    "startMs": 403320,
    "text": "And yeah, actually before I uh further,"
  },
  {
    "startMs": 406080,
    "text": "there's this new concept that we're"
  },
  {
    "startMs": 407400,
    "text": "introducing here called the prefix"
  },
  {
    "startMs": 408720,
    "text": "cache. So, the prefix cache is going to"
  },
  {
    "startMs": 411400,
    "text": "be looking at the"
  },
  {
    "startMs": 414480,
    "text": "um agent stream, and seeing if we"
  },
  {
    "startMs": 417160,
    "text": "already have generated audio for that"
  },
  {
    "startMs": 419720,
    "text": "sequence of words um from like a prior"
  },
  {
    "startMs": 423000,
    "text": "generation, or maybe like the same"
  },
  {
    "startMs": 424520,
    "text": "generation"
  },
  {
    "startMs": 425919,
    "text": "um in this"
  },
  {
    "startMs": 427840,
    "text": "uh in this call as as well."
  },
  {
    "startMs": 430000,
    "text": "So, um it sees the word you. Uh we for"
  },
  {
    "startMs": 433280,
    "text": "this prefix cache, we're going to be,"
  },
  {
    "startMs": 435800,
    "text": "you know, we don't want to like"
  },
  {
    "startMs": 437200,
    "text": "immediately hit on every single word."
  },
  {
    "startMs": 439320,
    "text": "We're going to be waiting for a little"
  },
  {
    "startMs": 440600,
    "text": "bit more words."
  },
  {
    "startMs": 442560,
    "text": "Um so, after three words, the prefix"
  },
  {
    "startMs": 445880,
    "text": "cache gets our first hit."
  },
  {
    "startMs": 447600,
    "text": "And um over here on the right, this is"
  },
  {
    "startMs": 450120,
    "text": "kind of our text-to-speech standard"
  },
  {
    "startMs": 452560,
    "text": "provider, you know, Cartesia is a"
  },
  {
    "startMs": 454160,
    "text": "text-to-speech engine with web socket"
  },
  {
    "startMs": 455960,
    "text": "support. So, we're we're piping the"
  },
  {
    "startMs": 458160,
    "text": "agent through the the cache, and also"
  },
  {
    "startMs": 461520,
    "text": "piping it through web socket."
  },
  {
    "startMs": 464840,
    "text": "Um more tokens come in, more cache, more"
  },
  {
    "startMs": 468040,
    "text": "sending through web socket. Not much to"
  },
  {
    "startMs": 469880,
    "text": "say here."
  },
  {
    "startMs": 471480,
    "text": "And okay, so now we get our first uh"
  },
  {
    "startMs": 475400,
    "text": "kind of first unique thing, which is we"
  },
  {
    "startMs": 478080,
    "text": "found the token that actually causes a"
  },
  {
    "startMs": 480480,
    "text": "cache miss. And it makes sense. If we're"
  },
  {
    "startMs": 482560,
    "text": "kind of caching previous generations,"
  },
  {
    "startMs": 485000,
    "text": "um you said your name is is a pretty"
  },
  {
    "startMs": 486720,
    "text": "common thing, but once you we add in the"
  },
  {
    "startMs": 488640,
    "text": "name, suddenly we're that's that's going"
  },
  {
    "startMs": 490960,
    "text": "to result in the cache miss."
  },
  {
    "startMs": 492720,
    "text": "At this point, we're actually going to"
  },
  {
    "startMs": 493880,
    "text": "yield out our cached audio. So, you said"
  },
  {
    "startMs": 496400,
    "text": "your name is is going to be"
  },
  {
    "startMs": 498720,
    "text": "um emitted as the rest of the streaming"
  },
  {
    "startMs": 501120,
    "text": "text is coming back. So, at this point,"
  },
  {
    "startMs": 503560,
    "text": "the user hears the agent and user"
  },
  {
    "startMs": 505800,
    "text": "doesn't really know what's going on."
  },
  {
    "startMs": 507080,
    "text": "They just looks like really fast"
  },
  {
    "startMs": 508680,
    "text": "response times to them."
  },
  {
    "startMs": 511760,
    "text": "Um and now the kind of remaining text"
  },
  {
    "startMs": 513800,
    "text": "flows through."
  },
  {
    "startMs": 515360,
    "text": "And at this point, we've already emitted"
  },
  {
    "startMs": 517560,
    "text": "from the cache. The cache has done its"
  },
  {
    "startMs": 519159,
    "text": "job. Um the rest we can kind of throw"
  },
  {
    "startMs": 522120,
    "text": "into Cartesia."
  },
  {
    "startMs": 523919,
    "text": "And here's kind of the trick where"
  },
  {
    "startMs": 527720,
    "text": "Cartesia has seen the entire transcript"
  },
  {
    "startMs": 531640,
    "text": "up to this point. It"
  },
  {
    "startMs": 534160,
    "text": "to to Cartesia, like it doesn't know"
  },
  {
    "startMs": 536320,
    "text": "about the existence of this prefix"
  },
  {
    "startMs": 537720,
    "text": "cache. It's just going to generate this"
  },
  {
    "startMs": 539960,
    "text": "full sentence with, you know, standard"
  },
  {
    "startMs": 541760,
    "text": "natural prosody."
  },
  {
    "startMs": 543480,
    "text": "But, what we do is when the generation"
  },
  {
    "startMs": 546440,
    "text": "comes back, since we've already played"
  },
  {
    "startMs": 548080,
    "text": "the audio here, we can actually suppress"
  },
  {
    "startMs": 549960,
    "text": "the audio from Cartesia here and just"
  },
  {
    "startMs": 552760,
    "text": "play out the remaining stuff. So,"
  },
  {
    "startMs": 555560,
    "text": "the user, there's might be a tiny bit of"
  },
  {
    "startMs": 558240,
    "text": "a hiccup. You know, I'll play some audio"
  },
  {
    "startMs": 560760,
    "text": "later and you'll know that you probably"
  },
  {
    "startMs": 562839,
    "text": "won't be able to notice."
  },
  {
    "startMs": 564320,
    "text": "And effectively, we're kind of taking"
  },
  {
    "startMs": 566520,
    "text": "this and"
  },
  {
    "startMs": 568800,
    "text": "or taking this part and just emitting it"
  },
  {
    "startMs": 571760,
    "text": "directly after this audio has finished"
  },
  {
    "startMs": 573920,
    "text": "playing. So, it looks seamless to the"
  },
  {
    "startMs": 575320,
    "text": "user."
  },
  {
    "startMs": 577080,
    "text": "And yeah, so I guess at this point kind"
  },
  {
    "startMs": 580560,
    "text": "of this is a lot of technical details."
  },
  {
    "startMs": 582760,
    "text": "I'll I mean the best way is just kind of"
  },
  {
    "startMs": 584560,
    "text": "play a call and see"
  },
  {
    "startMs": 587680,
    "text": "see this in action. So here I recorded a"
  },
  {
    "startMs": 590960,
    "text": "call and let's just hear what happens."
  },
  {
    "startMs": 595120,
    "text": ">> This is Bo OBGYN Elise speaking."
  },
  {
    "startMs": 600080,
    "text": ">> Hi, I think I might be pregnant and I"
  },
  {
    "startMs": 602240,
    "text": "want to schedule an ultrasound to"
  },
  {
    "startMs": 603880,
    "text": "confirm."
  },
  {
    "startMs": 606160,
    "text": ">> That's exciting news. Can you provide"
  },
  {
    "startMs": 608800,
    "text": "your name and date of birth?"
  },
  {
    "startMs": 612040,
    "text": ">> Sure, it's Elise Trial."
  },
  {
    "startMs": 614440,
    "text": "Date of birth"
  },
  {
    "startMs": 615880,
    "text": "2303."
  },
  {
    "startMs": 621680,
    "text": ">> Thank you. Are you a new patient with"
  },
  {
    "startMs": 623880,
    "text": "us?"
  },
  {
    "startMs": 626280,
    "text": ">> Yes."
  },
  {
    "startMs": 629240,
    "text": ">> Welcome to Bo OBGYN. Can I text you a"
  },
  {
    "startMs": 632440,
    "text": "link to upload your insurance"
  },
  {
    "startMs": 633760,
    "text": "information?"
  },
  {
    "startMs": 636920,
    "text": ">> Sure."
  },
  {
    "startMs": 639040,
    "text": ">> Okay, sent. Let me know when you've"
  },
  {
    "startMs": 641120,
    "text": "received it."
  },
  {
    "startMs": 648680,
    "text": "I've received your insurance"
  },
  {
    "startMs": 649839,
    "text": "information. The earliest appointment I"
  },
  {
    "startMs": 652200,
    "text": "have is Thursday, July 2nd at 10:00 a.m."
  },
  {
    "startMs": 655280,
    "text": ">> Actually, give me a second."
  },
  {
    "startMs": 657040,
    "text": "Let me check my calendar."
  },
  {
    "startMs": 661400,
    "text": ">> Sure, take your time."
  },
  {
    "startMs": 664480,
    "text": ">> Um do you have something for next week?"
  },
  {
    "startMs": 669160,
    "text": ">> For next week, I can offer an ultrasound"
  },
  {
    "startMs": 671400,
    "text": "with Dr. Avery Stone at North Clinic on"
  },
  {
    "startMs": 673880,
    "text": "Tuesday, July 7th at 2:00 p.m. or 3:00"
  },
  {
    "startMs": 677640,
    "text": "p.m. Do either of those work for you?"
  },
  {
    "startMs": 680880,
    "text": ">> Yeah, 2:00 p.m. works."
  },
  {
    "startMs": 683800,
    "text": ">> Great. Your appointment has been booked."
  },
  {
    "startMs": 686560,
    "text": "We look forward to seeing you then."
  },
  {
    "startMs": 689800,
    "text": ">> Thanks. Bye-bye."
  },
  {
    "startMs": 692839,
    "text": ">> All right. Yeah, that's pretty much it."
  },
  {
    "startMs": 694880,
    "text": "Um"
  },
  {
    "startMs": 695920,
    "text": "Yeah, you can kind of see our"
  },
  {
    "startMs": 698280,
    "text": "all this like uh"
  },
  {
    "startMs": 700720,
    "text": "streaming"
  },
  {
    "startMs": 702240,
    "text": "and you know, a lot of things are"
  },
  {
    "startMs": 703440,
    "text": "happening in the background and and you"
  },
  {
    "startMs": 704800,
    "text": "know, this is what really makes like"
  },
  {
    "startMs": 706360,
    "text": "voice agents interesting. And um there's"
  },
  {
    "startMs": 709240,
    "text": "a lot of effort that can be done in the"
  },
  {
    "startMs": 711320,
    "text": "harness to really kind of get a um"
  },
  {
    "startMs": 715280,
    "text": "natural conversation, which is what"
  },
  {
    "startMs": 717400,
    "text": "we're after."
  },
  {
    "startMs": 719440,
    "text": "Uh okay. Yeah, so I guess briefly, you"
  },
  {
    "startMs": 721800,
    "text": "know, in the last part, I want to just"
  },
  {
    "startMs": 724040,
    "text": "talk a little bit about Elise. So I"
  },
  {
    "startMs": 726520,
    "text": "think Elise, you know, our headquarters"
  },
  {
    "startMs": 728480,
    "text": "are in New York and kind of we're trying"
  },
  {
    "startMs": 730080,
    "text": "to expand our presence here in the Bay"
  },
  {
    "startMs": 731760,
    "text": "Area. Um we I think it's maybe like a"
  },
  {
    "startMs": 735280,
    "text": "different style of company that I think"
  },
  {
    "startMs": 737640,
    "text": "people are"
  },
  {
    "startMs": 739640,
    "text": "like uh think of when they think about"
  },
  {
    "startMs": 741000,
    "text": "AI startups in San Francisco. Where"
  },
  {
    "startMs": 743800,
    "text": "we're actually very focused on um"
  },
  {
    "startMs": 746680,
    "text": "just like helping people and helping"
  },
  {
    "startMs": 750360,
    "text": "people where they need it, like kind of"
  },
  {
    "startMs": 752040,
    "text": "the life's most critical areas. We work"
  },
  {
    "startMs": 753920,
    "text": "on housing, health care and uh we're"
  },
  {
    "startMs": 757520,
    "text": "doing really well and you know, here's"
  },
  {
    "startMs": 760440,
    "text": "there's a link here to"
  },
  {
    "startMs": 762680,
    "text": "um you kind of join our team and there's"
  },
  {
    "startMs": 764640,
    "text": "going to we're going to be uh posting a"
  },
  {
    "startMs": 766680,
    "text": "lot on Twitter, so you can follow us at"
  },
  {
    "startMs": 768360,
    "text": "EliseAI as well."
  },
  {
    "startMs": 770600,
    "text": "Um yeah, that's that's it."
  },
  {
    "startMs": 773563,
    "text": ">> [applause]"
  }
]