Eshwary Mishra
Machine learning · High energy physics · Low-latency C++ · Game engines
Writings
Notes on tokenizing hadronic collision events, sequence packing, validation, starter prompts, and evaluating generative models for high-energy physics.
9 min read 02 What game engines taught me about latency budgetsA personal engineering note connecting Unreal Engine work, high-performance C++, backend services, profiling, and design under hard real-time constraints.
7 min readProjects
particleGPT
A GPT-style modeling pipeline for particle collision events: tokenization, training, sampling, untokenization, validation, metrics, and analysis.
Scaling laws visualizer
A technical design tool for reasoning about compute, data, model size, and expected loss from scaling-law papers.
Realtime gameplay systems
Gameplay, networking, rendering-adjacent, and desktop tooling work built around performance, correctness, and iteration speed.
Cloud-backed applications
Backend systems using GameLift, RDS, DynamoDB, Cognito, Lambda, SNS, S3, and related infrastructure pieces.
HEP metrics lab
Placeholder project for generative-model evaluation, detector-inspired plots, and comparison metrics for physics event data.
Tokenizer workbench
Placeholder project for experimenting with event-stream tokenization, starter prompts, sequence packing, and reversible data transforms.
Latency notebook
Placeholder project for profiling notes, frame-time experiments, memory behavior, and low-latency systems engineering patterns.
Research infra atlas
Placeholder project for backend architecture, AWS infrastructure, experiment tracking, data movement, and research tooling systems.