Medium

Manjunath Janardhan

In

Mac O’Clock

by

Manjunath Janardhan

·

Oct 23

Apple Says Its Built-In AI Isn’t for Coding. I Gave It 13 Coding Jobs on a 16 GB Mac mini Anyway.

macOS 27’s new fm command runs Apple’s on-device model free and offline. It couldn’t drive Opencode, but with Aider it fixed bugs and added…

Apple Says Its Built-In AI Isn’t for Coding. I Gave It 13 Coding Jobs on a 16 GB Mac mini Anyway.

In

Mac O’Clock

by

Manjunath Janardhan

·

Oct 22

I Turned My 16 GB Mac mini Into an Offline Coding Agent With Recursive’s Flint

Recursive says Flint needs a 48 GB Mac. Unsloth’s 11 GB build fixed a broken project on my mini in every run I gave it.

I Turned My 16 GB Mac mini Into an Offline Coding Agent With Recursive’s Flint

In

Mac O’Clock

by

Manjunath Janardhan

·

Oct 20

I Cleaned Up My YouTube Feed With a Free AI That Runs Fully Offline on My Mac

A small AI model inside Chrome hides the YouTube videos you don’t want, almost as accurately as TypeSafe’s cloud model Jev, but nothing you…

I Cleaned Up My YouTube Feed With a Free AI That Runs Fully Offline on My Mac

In

Data Science Collective

by

Manjunath Janardhan

·

Sep 29

I Built an Open Rival to TypeSafe’s Jev. Here’s Where It Wins and Where It Doesn’t.

From a 400M model that runs on a CPU to a 30B that beats Jev on unseen decisions: same questions, same scorer, every answer public. Plus…

I Built an Open Rival to TypeSafe’s Jev. Here’s Where It Wins and Where It Doesn’t.

In

Data Science Collective

by

Manjunath Janardhan

·

Sep 26

An Open-Source Rival to TypeSafe’s Jev Runs on a 16 GB Mac.

Laya answers in 20 to 50 milliseconds on a Mac, for free. It claims to beat Jev, but its launch compared it with Jev numbers someone else…

An Open-Source Rival to TypeSafe’s Jev Runs on a 16 GB Mac.

In

Data Science Collective

by

Manjunath Janardhan

·

Sep 25

I Measured Prism ML’s Bonsai 2 Against My Own Builds of Alibaba’s Qwen3.8–27B.

36 tokens per second in about 10 GB on an M4 Max, the same test scores as a 4-bit build twice its size, and five traps that nearly skewed…

I Measured Prism ML’s Bonsai 2 Against My Own Builds of Alibaba’s Qwen3.8–27B.

In

Mac O’Clock

by

Manjunath Janardhan

·

Sep 21

Run Prism ML’s Bonsai 2 on a 16GB Mac mini: Alibaba’s 27B Qwen 3.8 in 8.6 GB, Fully Offline

Bonsai 2 squeezes Alibaba’s 27-billion-parameter Qwen3.8, which reads text and images, into 8.6 GB. Here’s the setup, what it’s like on the…

Run Prism ML’s Bonsai 2 on a 16GB Mac mini: Alibaba’s 27B Qwen 3.8 in 8.6 GB, Fully Offline

In

Data Science Collective

by

Manjunath Janardhan

·

Sep 19

I Tested TypeSafe’s Jev, a 478× Cheaper Decision Model, Against Claude, GPT-6, Kimi, MiniMax and…

TypeSafe’s Jev answers typed questions in under half a second for $0.025 per thousand. I gave it and five frontier LLMs the same 200…

I Tested TypeSafe’s Jev, a 478× Cheaper Decision Model, Against Claude, GPT-6, Kimi, MiniMax and…

In

Coding Nexus

by

Manjunath Janardhan

·

Sep 17

Is AGI Here? Analyzing Jensen Huang’s Bold Claim on Scaling, Chips, and Autonomous Agents

A Deep Dive into OpenAI’s GPT-6 Astra, Custom Silicon, and the Paradigm Shift for Software Engineers

Is AGI Here? Analyzing Jensen Huang’s Bold Claim on Scaling, Chips, and Autonomous Agents

In

Mac O’Clock

by

Manjunath Janardhan

·

Sep 16

Run Alibaba’s Qwen Flash-Next 180B Locally on a Mac 64GB: 50 Tokens a Second, Fully Offline

Qwen’s 180-billion-parameter Flash-Next answers at about 50tokens a second on a 64 GB Mac, completely offline. Here’s the exact setup, with…

Run Alibaba’s Qwen Flash-Next 180B Locally on a Mac 64GB: 50 Tokens a Second, Fully Offline

Manjunath Janardhan

Manjunath Janardhan

AI/ML Computational Science Senior Manager at Accenture with 21+ years building enterprise AI, GenAI, and intelligent operations.

Following

Help

Status

About

Careers

Press

Blog

Store

Privacy

Rules

Terms

Text to speech