Summer 2026 ML Systems Intern

Partcl is the vibe coding platform for chip designers.

I was told to make this job description “sound cool and not corporate” so I had ChatGPT translate it into gen alpha:

EDIT: We had an overwhelming number of applicants, so we’re having everyone do a short take home challenge and we’ll interview the people who do well on this: https://github.com/partcleda/intern_challenge

About Partcl

We’re cooking the next wave of chip‑design automation—built for speed, scale, and builder productivity. We believe AI should level up hardware engineering, and the first W is smarter optimization tools.

If you want to tackle massive‑scale problems in physical AI, Partcl is the move.

What you will do:

• Cook + train custom models to go after NP‑complete optimization problems (smart heuristics that actually ship, fr)
• Spin high‑perf CUDA kernels for training + inference and make them zoom
• Speedrun parsers that chew through massive files without choking
• Build ultra‑lean, cache‑friendly data structures — minimal mem, max throughput
• Hook up an LLM interface to those structures so users can literally talk to their design

What you bring:

• Python + PyTorch skills from classes or projects; you’ve trained & evaluated custom models (not just “wrapped a GPT,” fr)
• Systems intuition: you grok memory hierarchies, latency vs. throughput trade‑offs, & why cache misses make everyone sad
• Some CUDA exposure or you’re dangerously motivated to pick it up fast—kernels don’t scare you
• Range: you can bounce between model design, low‑level perf work, and scalable system plumbing (think N → 10M+)
• Tooling chops: you’ve at least dabbled in GPU profiling/perf analysis and you’re down to go deeper

Bonus XP (nice to have)

• Any exposure to EDA / chip design flows
• Reinforcement Learning beyond toy notebooks
• Research or serious projects in compilers / programming languages / databases
• Publications in respected venues (conference/journal/tech reports)
• Physics‑driven ML or numerics that make simulators blush

Why this rocks

• Work on problems where milliseconds matter and terabytes aren’t “big,” they’re Tuesday.
• Ship code that touches real silicon workflows.
• Learn from builders who live at the ML × systems × hardware edge.

Pull up with your resume/portfolio/GitHub. Let’s build.