Kunal Batra

Developer Relations @Datadoghq. Previously @AWS, @Clarifai, @Auth0, @SendGrid Opinions are my own.

[Jeremy @linuxquestions]

The @datadoghq DASH fireside chat lineup looks amazing.

@oliveur + @ClemDelangue
Alexis Lê-Quôc + @thsottiaux
@atalwalkar + @_sholtodouglas
Yanbing Li + @tomocchino

I absolutely can't wait. See you in June!

https://t.co/PfMHEm4qP9

[Kunal Batra @kunal732]

👀

[Shashank Prasanna @shshnkp]

Cool new agent skill for agent assisted conversion of transformers models to MLX! 🔥
Blog post by @awnihannun and @pcuenq

https://t.co/ld7TfWcrtJ

[Ameet Talwalkar @atalwalkar]

Heading to @iclr_conf next week? Datadog AI Research is hosting an event along with the ICLR SPOT workshop organizers. Join us to meet our team and learn about our work on observability world models + agentic post-training!

[Kunal Batra @kunal732]

Been working on this for a bit, stoked to announce a community port of the AWS Strands Agents framework to Swift. It can run natively on device and you can build those fun OpenClaw-style computer use examples. It can run inference locally via MLX or through Bedrock.

In the video I connected an agent to a Mac MCP server and asked it to open Word and write a sentence. It also sends OTel telemetry to your tool of choice. In the video you can see the telemetry from the agent being shown in Datadog's LLM Observability.

https://t.co/bd4qc4uB1o

Coming soon: Interacting with the agents through voice either AWS Nova Sonic or locally with @Prince_Canuma's mlx-audio-swift package.

cc: @awsdevelopers, @awnihannun , @shshnkp

[Kunal Batra @kunal732]

Introducing MLX-Swift-TS
https://t.co/TDCJXVpago
An SDK for running time series foundation models fully on-device on Apple Silicon.

When I joined @datadoghq , I was introduced to Toto, our time series foundation model, and got excited about zero-shot forecasting across different domains. While building a health copilot app, I realized there wasn’t a simple way to run models like these locally on device. So I built one.

MLX-Swift-TS exposes a common TimeSeriesForecaster interface for loading and running multiple time series architectures directly in Swift using MLX. No server required.

The attached video shows on-device forecasting running inside a native Swift app.

Huge thanks to @awnihannun and the MLX team for building MLX and its Swift API, @Prince_Canuma for inspiration on MLX SDK patterns, and @atalwalkar and the Datadog team for Toto.

[Kunal Batra @kunal732]

@mgonto Do you have any frameworks you prefer for constructing/running these ?

[Kunal Batra @kunal732]

@fullstackpython Ha fair - “almost” does a lot of work there 😉
Always appreciate a good meme though.

[Kunal Batra @kunal732]

Always enjoy chatting with this crew

[Kunal Batra @kunal732]

@goosewin @RhysSullivan Ha

[Kunal Batra @kunal732]

@shshnkp @awnihannun @angeloskath This is great!

[Kunal Batra @kunal732]

@crtr0 same ✉️

[Kunal Batra @kunal732]

@Prince_Canuma Mlx evangelist !!

[Kunal Batra @kunal732]

@amit Nice! I see you got the noise canceling ones

[Kunal Batra @kunal732]

@Prince_Canuma Do you have or know any resources to help convert times series foundation models like google's timesfm to MLX ? From my understanding these use patch embeddings for inputs in a similar way to how image inputs are encoded for vlms .

[Kunal Batra @kunal732]

@sach1n @Prince_Canuma @awnihannun Thank you! Have been spinning my wheels the past couple of days to get 7B-Instruct to work but keep running into issues

[Danilo Poccia @danilop]

Great work by @kunal732 to help run locally using MLX the @arcee_ai Virtuoso Lite 10B model distilled from DeepSeek V3.

[Kunal Batra @kunal732]

@mkurman88 @reach_vb This is great! Thank you!

[Kunal Batra @kunal732]

@vatsal_manot cc: @shshnkp - not sure if there is an internal logging system for feedback like we had at AWS but for developers. But something to note.

[Kunal Batra @kunal732]

@vatsal_manot It's what makes MLX surprising. Just doesn't feel like its from apple.