Apr 6, 2024 · 58m · latent-space
Personal AI Meetup - Bee, BasedHardware, LangChain LangFriend, Deepgram EmilyAI
Harrison Chase · 10m spoken Ethan Sutin · 10m spoken Damien Murphy · 8m spoken Shawn Wang · 7m spoken Nik Shevchenko · 7m spoken Deepgram Aura AI Assistant · 33s spoken Pirate Voice · 12s spoken
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
This meetup covers the tripartite foundation of personal AI companions, showcasing innovations in low-latency voice pipelines, open-source wearable hardware for continuous life logging, and long-term memory software architectures.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The hosts as informed peer 3.8 Guest teaching 4.2 Guest disagreement 0.0 The hosts pushing back 0.3
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 30:59 Ethan contrasts privacy perspectives on wearable recording
In a meetup marked by harmony, Ethan offers the only notable disagreement by rejecting Igor's surreptitious smartphone camera method in favor of transparent recording.
Hardest push from the hosts ▶ 52:00 Shawn probes memory operational semanticsShawn pushes past basic memory retrieval frameworks to ask whether operational semantics can be decoupled from raw prompt insertion.
Biggest teaching moment ▶ 18:05 Damien breaks down API latency nuancesDamien educates the room on the significant second-level latency spikes of OpenAI's hosted endpoints versus deploying dedicated instances on Azure.
The host holds their own ▶ 56:22 Shawn proposes RAPTOR for memory systemsShawn demonstrates sharp technical insight by suggesting the RAPTOR paper's hierarchical clustering approach for memory consolidation, which Harrison admits he had not thought of and praises.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Shawn Wang Demonstrates Live Vapi Voice AI Assistant | 0 | 0 | 0 | 0 | Shawn kicks off the meetup with an introductory demonstration creating a pirate-themed live voice AI assistant using Vapi. Because this is an opening host monologue and live demo, interactive host-guest dynamics are at zero. | |
| Igor Presents Continuous 24/7 Life Audio Recording Setup | 5 | 2 | 0 | 0 | Igor presents his setup for continuous 24/7 life audio recording, and Shawn contributes his own iOS shortcut workflow for automated meeting transcription. The exchange is completely collaborative with both sharing workflow tips. | |
| Damien Murphy on Building Ultra Low-Latency Voice Bots | 4 | 6 | 0 | 1 | Damien delivers a technical walkthrough on building ultra-low-latency voice bots across STT, LLMs, and TTS pipelines. Shawn asks targeted questions regarding GPT-3.5 versus GPT-4 latency trade-offs. | |
| Ethan Sutin Demonstrates Wearable AI and Agent Actions | 4 | 5 | 0 | 0 | Ethan demonstrates wearable continuous audio capture alongside autonomous agent actions controlling mobile apps. Shawn asks focused questions clarifying the cloud browser streaming architecture. | |
| Nik Shevchenko on Open Source Wearables and Friend Hardware | 3 | 5 | 0 | 0 | Nik outlines the open-source wearable landscape and hardware challenges on low-power chips. Shawn prompts Nik to explain how they achieved significant voice quality improvements within constrained onboard memory. | |
| Harrison Chase on AI Memory Systems and LangChain Journaling | 7 | 7 | 0 | 1 | Harrison presents LangChain's memory abstractions and journaling architecture, leading into a technical discussion with Shawn on memory decay and RAPTOR hierarchical summarization. Both host and guest display substantial technical depth. |
Statements from this episode (9)
Wang plans default-recording, opt-out policy for June AI Engineer conference.
“Something I want to do for my conference in June is, like, everything should be default recording, and then you opt out instead of opting.”
Murphy: Voice bot latency over 1.5 seconds triggers users to repeat themselves.
“So essentially, if you go beyond, say, 1.52 seconds, a lot of people will actually say something again, right? They think that the person is no longer there on the other end.”
Murphy: Azure OpenAI significantly beats OpenAI's hosted API 400-600ms latency.
“And then GPD, 3.5 turbo or four, you probably get, you know, 400, maybe 600 milliseconds of latency in their hosted API. And if you go into Azure and you use their services, you can get that down a lot lower.”
Murphy: Five-minute voice calls cost 6.5 cents on Deepgram versus ElevenLabs.
“And then on the text-to-speech side, and doing something like this with an 11 labs would be about maybe a dollar 20. And just to give you an idea of comparison. So you can do a five minute call here for about six and a half cents.”
Sutin: Local model setup friction killed Owl AI's open-source developer adoption.
“I learned, like, we did not make the developer experience very good. It was very complicated like, because we were using, like, local whisper, local models, and, like, getting it to work on CUDA, Mac, Windows. We didn't do a good job, so it was very difficult …”
Sutin: AI wearables require Bluetooth, contrary to Humane and Rabbit's LTE approach.
“I think Humane and Rabbit are both LTE and Wi-Fi, but to get like a wearable, you really need Bluetooth.”
Chase: Nobody in the AI industry knows how to properly solve memory.
“I don't think anyone knows how to deal with memory, and so I think all these different approaches are...”
Chase: Long-term LLM memory systems must incorporate mechanisms for memory decay.
“Yeah, I think there absolutely needs to be some sort of memory decay, or some sort of, like, invalidating previous memories.”
Chase: LangChain's experimental AI journaling app struggles to consolidate redundant memories.
“So one of the issues that we haven't really tackled is in this journaling app, if you notice in here, there's a bunch of ones that are really similar, right? And so, like, there's a clear kind of, like, consolidation or update or something procedure that, that…”