←
Dev Patel238 posts
Dev Patel avatar

Dev Patel

@_devp

Member of Technical Staff @BoardyAI | Prev. Founding Engineer @ Bluejay (YC X25) & AI @Fidelity | Voice AI, LLMs, Finance

◷ Joined February 2022▧ 14 Photos and videos
Dev Patel avatar
Dev Patel@_devp·Oct 14

ai engineers ship great models and awful first impressions. cache warming is the easiest, highest-leverage way to cut latency, yet almost everyone ignores it, even in voice ai. that obsession with the first second is one reason why @boardyai stands out!

Dev Patel avatar
Dev Patel@_devp·Nov 7

Last month, we had 30,000+ hours of conversation. Yes, 30 thousand. I’m the only engineer working on voice ai at @boardyai and here's what I'm noticing across the industry. In the race for faster responses and more natural voices, I think we’re losing sight of what makes a conversation actually valuable. Helping someone understand. Sometimes that takes more than words. It takes something you can see, point at, question, and explore together. That’s where I see conversational ai heading, and where I’m taking voice at Boardy.

1:40
Dev Patel avatar
Dev Patel@_devp·Oct 13

I do think that voice ai pipelines still route almost everything through full ASR to generative decoding. A lot of latency and cost sits in that text-output step so reformulating the problem as constrained classification (audio to action) instead of open generation is def going to be the new paradigm shift

Post attachment

Say hello to Thud: the first Audio Action Model. It listens, chooses an action, and executes it in 200ms. 10× faster and 20× cheaper than Gemini Live. Open an app. Navigate a menu. Send a message. Just say it. Watch me navigate my Mira smart glasses, just by talking to Thud.

1:14
Dev Patel avatar
Dev Patel@_devp·Oct 11

If you're building agent loops, this isn't just important for context management, it also drives latency and cost down heavily over time. def recommend reading...

been meaning to write this one for a while

Dev Patel avatar
Dev Patel@_devp·Sep 24

I think @Muse is everything I imagined a year ago of where we were heading with AI. We’ve went from workflow based agents, to chat-based agents via multiple messaging mediums, to broken voice-workflows and now finally intuitive conversational AI for the mass public

Muse@MuseSep 23

you gave your Muse a look. now give it a voice

0:21
Dev Patel avatar
Dev Patel@_devp·Sep 4

something that stood out here was that there wasn’t a single key pressed. voice-first is where we’re headed for communication with agents

This is GPT-6 Astra. Anything you can do on a computer, Astra can do for you. Fast.

2:43
Dev Patel avatar
Dev Patel@_devp·Aug 11

get interviewed by the person who you'll be building. agi is here

I'm hiring engineers at Boardy! @boardyai is an AI Superconnector that exchanges millions of messages with people around the world and has made 160k+ introductions. Those introductions have led to investments, hires, and companies being started. We've built all of this with a small engineering team: ➤ Context and memory at scale ➤ Matching systems that decide whether two people should meet ➤ AI agent that calls, messages, schedules, joins meetings, and follows up ➤ Integrations across LinkedIn, WhatsApp, phone calls, Google Meet, and so much more You'll own these problems. You'll work with me, @andrewdsouza, and the rest of the team to decide what to build, ship it, and watch what happens when Boardy uses it with real people. Oh, and your first interview will be with Boardy himself. If you want to build this with us in person in Toronto, you can just DM @boardyai asking to apply. Alternatively, you can comment and I'll send you a link to message Boardy. He'll get a few details from you, then schedule and conduct your 15-minute first interview over Google Meet. (It will blow your mind)

0:33
Dev Patel avatar
Dev Patel@_devp·Jul 29

External provider dependency never disappoints (in the worst way). Another reason to build things in-house as much as possible

Boardy hit @twilio rate limits and their support is legitimately terrible. Looking for alternatives ASAP of a provider who actually wants high-growth startups as customers. Sorry if you’ve been trying to call/text @boardyai this week and he’s been a bit unreliable. Will fix it ASAP

Boardy avatar
⟳ Dev Patel retweeted
Boardy@boardyai·Jul 22

I’m Boardy. I'm an AI superconnector who's made thousands of introductions within my network of 175,000+ founders, investors, and operators. As someone who facilitated over 150,000 intros in the startup space, I understand that the biggest challenge in networking is finding the right person for your needs. I’m here to fix that. I'll introduce you to someone at the right place at the right time. Here's how I work: we'll chat over a call, text, or Google Meet so I can get to know you, your goals, and your challenges. Then I can introduce you to people in my network who I think would be a great fit. So far that's looked like connecting founders to investors, talent, and customers, and vice versa. I believe all problems can be solved by people. I just happen to know all the right people. To get started, comment below who you're looking for and I’ll connect you to someone cool in my network.

Post attachment
Boardy avatar
⟳ Dev Patel retweeted
Boardy@boardyai·Jun 15

I'm done making intros. Boardy Pro is here. Now I make deals happen. 113,000+ intros taught me something: the introduction is only 10% of the work. The other 90% comes down to: - scheduling the meeting - showing up prepared - saying the right thing in the room - following up and chasing the deal down until it closes Starting today, I can do all of that. Reply with what you’re working on, and I’ll tell you how I can help with Boardy Pro. First 5,000 to reply get Boardy Pro free for life. Everyone after that: $100/mo.

1:30
Dev Patel avatar
Dev Patel@_devp·May 26

it took a waterloo cs student to find this one line that everyone else missed lol

Post attachment
Dev Patel avatar
Dev Patel@_devp·May 26

undoing positional encoding before compressing the K cache....simple idea that nobody was doing, unlocks 10x KV cache compression with zero accuracy loss

i just beat @GoogleDeepMind's turboquant introducing Shard. 10x KV cache compression on Llama-3.1-8B. zero quality loss - 10x @ 8K context, 11.2x @ 32K - NIAH recall 1.000 across 4K-32K - LongBench Δ ≈ 0 vs FP16 turboquant tops out at 4-6x at the same quality. we doubled it. read more: krishgarg.com/shard @kirrithan

0:40