Yash Khurana

AI Engineer. Making local AI actually usable →

[Yash Khurana @yolo_yash_]

Hi, I'm Yash.

I'm building OvertChat — the front door for local AI. Your models, your server, your family's accounts. No API keys, no cloud.

Because a model on your GPU is just infrastructure. Local AI that's actually usable → https://t.co/h31XEPjXr1

[Yash Khurana @yolo_yash_]

@MiaAI_lab I don't they don't pull a deepseek on us. Dsv4 fit perfectly in 2x pro 6000s. 4.1 needs 4x lol.

[Yash Khurana @yolo_yash_]

@djnmrjnvc @TechMDAI Yeah, just couldn't find a recipe for it. Made one myself in case anyone else needs it - https://t.co/AWL4YtQp5U (uses diffbots weights)

[Yash Khurana @yolo_yash_]

@TechMDAI Not everyone in the comments simping for your PRO 6000s🥀

[Yash Khurana @yolo_yash_]

@noozie @TechMDAI Tried Maya, it's definitely good but no parallel decode kills it for me. Requests get queued unlike GLM 5.3F on Maya. Maybe they'll have a fix for it soon?

[Yash Khurana @yolo_yash_]

@TheDesertLynx Hell yeah, what can I add?

[Yash Khurana @yolo_yash_]

Hi, I'm Yash.

I'm building OvertChat — the front door for local AI. Your models, your server, your family's accounts. No API keys, no cloud.

Because a model on your GPU is just infrastructure. Local AI that's actually usable → https://t.co/h31XEPjXr1

[Yash Khurana @yolo_yash_]

@vanveluwen @arpitatscale @NamanyayG That's diabolical lmfao

[Yash Khurana @yolo_yash_]

@axitek @brianartigas To me it feels more like adult legos lol. I've never had more fun than tinkering with GPUs and assembly

[Yash Khurana @yolo_yash_]

@axitek @ai_christianson

[Yash Khurana @yolo_yash_]

@gabu3d_pl @gospaceport me running DavidAU/Qwen3.8-Flash-Next-Abliterated-Uncensored-Turbo-Brilliance-50x-Reasoning-Instruct-GSQ-RCO-GGUF on a single 3090

[Yash Khurana @yolo_yash_]

@NoemiTitarenco Ah yes, freedom

[Yash Khurana @yolo_yash_]

@axitek I'm more of a club 3090 guy myself (https://t.co/V4ZWXjHhjW). Running 3.8FN for now, until 4 drops!

[Yash Khurana @yolo_yash_]

Zero local AI experience. Stumbled on a random YouTube channel talking about why local matters, stop relying on the cloud so much. Saw the quad 3090 rig on a DIY rack and that gave me all the permission I needed.

My version just involved a lot of zip ties and 10x 3090s.

That's where I landed. No API. Big electricity bill. My whole household chats on it now.

@gospaceport thank you for the rabbit hole 🫡

[Yash Khurana @yolo_yash_]

@axitek Here you go, at one point it was 10x 3090s haha. It ran stable. Did not face any issues.

[Yash Khurana @yolo_yash_]

@axitek You don't necessarily need 240V circuit either haha. I was using 1x 1600W per circuit (I was using a 25ft extension cord for the second). It might sound janky, but if you don't have kids/pets in the house it works pretty well.

[Yash Khurana @yolo_yash_]

@gospaceport The local ai guy literally cooked!

[Yash Khurana @yolo_yash_]

@SamVale29 Actually good local AI - https://t.co/h31XEPjXr1

[Yash Khurana @yolo_yash_]

@moiiikaaa You're right to push back on that. It's not AI — it's SI.

[Yash Khurana @yolo_yash_]

@moonpetal76 vibe coding