Aaditya Mishra
computer engineering @ NC State
At Cast & Crew, I train an in-house transformer and build the evaluation system that catches regressions before they ship. At NC State's CoMAIL lab, I backpropagate through fluid simulations to recover physical parameters from noisy data.
Selected work
I trained a Unitree G1 in simulation to pick objects off the floor with PPO and no demonstrations. After the first policy gamed my success metric, I rebuilt the evaluator around contact physics; the new policy succeeded in 229 of 300 trials that began near the object. The full walk-to-grasp chain remains unsolved.
Python · PyTorch · Isaac Lab · RSL-RL
Curious attaches an 83M-parameter side transformer to a frozen 4B code model. I built the model and training system; the latest completed run reduced held-out next-token loss by 5.6% without changing the base weights, but it has not improved end-to-end coding yet.
Python · PyTorch · Hugging Face Transformers · CUDA
Gilfoyle wraps the Claude Code CLI in a cross-platform desktop app with persistent terminals, project switching, agent status, and Git diffs. Moving session indexing into Rust cut a 1,848-session benchmark from 2.8 seconds to 0.7.
TypeScript · Electron · React · Rust
I shipped NewsNine to the App Store and built the pipeline behind it: ingest coverage from more than 30 publishers, cluster overlapping stories, and produce one multi-source brief per event.
TypeScript · React Native · Expo · Supabase
CacheForge adapts the method from Google DeepMind's FunSearch paper to discover CPU cache algorithms. An LLM generates C++ replacement policies, compiles them into ChampSim, and benchmarks them on SPEC CPU traces; measured IPC and a retrieval memory of prior experiments guide each new generation.
Python · C++ · ChampSim · SQLite · Groq · GPT-4o
Experience
Cast & Crew · AI engineering intern
I work on document extraction used to process roughly 15,000 payrolls each week. I built an AWS service that tests each change before it merges, plus a CLI and Claude Code skill that trace failed files across S3, databases, and Kafka. I'm now training an in-house extraction model.
CoMAIL Lab, NC State↗ · undergraduate researcher
I built a differentiable fluid simulator in JAX that uses automatic differentiation and L-BFGS to fit physical parameters from pressure and flow data. It ran 2.4× faster than the previous solver on the same benchmark and showed which parameters could be recovered from noisy measurements. We're now moving to patient data.
Nutanix · systems engineering intern
I built search and indexing for SE-GPT, Nutanix's internal assistant for hybrid-cloud systems engineers. For the intern hackathon, I built local middleware that masked sensitive data before LLM requests left the user's machine; our team placed third.
SolarPack · programming lead
I lead software for NC State's solar race car. I integrated live CAN data into the driver display and fixed braking, battery, and drivetrain firmware used in competition. One of those fixes restored the brake light.
Outside work, I play poker.