Muhammad Siddique

20 • UT Dallas CS Alum • Prev - @amazon AGI

[Muhammad Siddique @mdsdq]

Five years ago I built my first gaming PC with an NVIDIA RTX 3060 Ti GPU. I had no idea fast forward to today, it would become my primary tool for running AI experiments, albeit a little painfully with only 8GBs of VRAM.

That's why I was excited to dive into the Google Cloud × NVIDIA learning pathways.

I've used the NVIDIA NIM API for inference in my projects but I've never actually deployed a NIM model myself, locally or on something like GKE. Walking through the full deployment flow was very informative, especially because I want to build future projects without always relying on APIs for LLM inference. The inference pathway solidified concepts I had intuitions about but never applied. The tradeoff between latency and throughput, the knee point, why decoding without a KV cache creates bottlenecks. Things for me to consider when evaluating the performance of my own deployed models.

The data analytics module was my favorite and probably the most applicable to me. I recently trained a hybrid Transformer-Mamba model on my RTX 3060 Ti for spatiotemporal gesture detection using the Jester dataset which has roughly 30 GB of data. Data preprocessing was a bottleneck I just accepted due to the massive size. Turns out cuDF accelerates pandas with zero code changes and I probably left a lot of performance on the table.

Going forward, I want to build adaptive AI agents that run locally on models I train myself, experimenting across Transformers, SSMs, and eventually world models. Understanding how to optimize inference and deploy models end-to-end is foundational to that goal, and these pathways gave me a much clearer picture of what that stack looks like in practice. As Jeff said in one of the videos, the best way to understand things is to try them. So that's the plan, keep experimenting on my 3060 Ti, Colab, and hopefully on a DGX Spark soon ;)

#NVIDIAGTC @googlecloud @googledevs

[Muhammad Siddique @mdsdq]

I built Vizpath: observability + intelligence + data curation for AI agents locally.

My thesis is as open models keep getting better, developers will run complex agents locally and want to finetune models specifically for their workflows.

Vizpath captures agent traces, auto-analyzes quality, and exports curated training data whilst all running locally and powered by Nemotron models which are excellent at data curation.

💻 GitHub: https://t.co/xMFFRtKxOe

#NVIDIAGTC @ctnzr

[Muhammad Siddique @mdsdq]

My workflow has been to run Claude & Codex in parallel to find bugs and potential enhancements, then feed all bugs to Claude to generate 2 separate fix plans (one for Claude, one for Codex) with clear separation, and assign each agent its plan to implement

[Muhammad Siddique @mdsdq]

been using gemini cli for a while.

just added claude code + codex to the mix.

going to use all three extensively and share how they compare.

[Muhammad Siddique @mdsdq]

I probably wasn’t using Gemini CLI as best as I could but uncovered a lot of bugs in existing projects with codex and claude code.

[Muhammad Siddique @mdsdq]

Grateful for an eventful & life changing year. Excited for 2026🎊

[Muhammad Siddique @mdsdq]

2025 Recap 📸

Jan–April: Took a gap semester from uni to recharge. Spent most of it in my home country, Pakistan, but also some time in the UAE and Saudi. 10/10 experience. Highly recommend taking a gap sem sometime before graduation.

[Muhammad Siddique @mdsdq]

August–December: Returned to Dallas to complete my final semester at UTD. Took interesting courses in the ML/AI space: Computer Vision, NLP, and Data Science. In the end, graduated with a BS in CS 🎓

[BOSS @thebeautyofsaas]

https://t.co/9rOTItoSUu

[Muhammad Siddique @mdsdq]

@mert Perplexity's deep research has been solid for me, especially with project planning. Not sure if other companies have better deep research features though

[Muhammad Siddique @mdsdq]

Hi, I’m Muhammad 👋

I’m a recent CS grad broadly passionate about AI/ML but most recently exploring AI agents.

Excited to share bits & pieces of my journey, learn from this community, and connect with fellow devs!