SwarmGPT
Language meets motion.
Turning natural-language instructions into drone swarm choreography with safe motion planning.
Software engineer & researcher
Building intelligence.
Engineering reality.
I'm Vedant. I build software at Arista Networks and explore the space where high-performance systems, robotics, and machine learning meet.
01 / selected work
Ideas, put to work.
Gazprea Compiler
From source to execution.
Built an end-to-end compiler for Gazprea, a full programming language with lexical analysis, parsing, semantic analysis, type checking, and code generation. In testing, our implementation outperformed the teaching assistant's reference solution. The source stays private under course policy because the assignment is reused in future offerings.
02 / profile
Curious about the layer between an idea and a working system.
I am Vedant, a computer scientist who enjoys moving between abstraction levels: from compiler IRs and network silicon to learning systems and the physical world. My work tends to be practical, research-minded, and shaped by a simple question: what would make this robust enough to use?
I graduated from the University of Alberta with a B.Sc. Honors in Computer Science, and I am happiest in teams that value clear thinking, careful experiments, and software that earns its complexity.
01Make the boundary clearGood interfaces let ideas travel.
02Measure the real thingPerformance is a property of context.
03Leave a useful trailResearch is better when it is legible.
03 / experience
Work history
A timeline of building, investigating, and teaching.
Jul 2025 - now
Software Engineer
Arista Networks · Vancouver, BC
- Develop and validate low-level software for ASIC and FPGA-based high-performance switching platforms, including congestion monitoring for Broadcom Tomahawk 6 ASICs.
- Migrated package builds to Bazel, reducing build times by 50% and improving developer iteration speed.
Jun 2024 - Sep 2024
Research Assistant, Dynamical Systems & Learning Lab
Technical University of Munich · Munich, Germany
- Building a framework and method to use LLMs to control a swarm of drones for choreography applications using Crazyflie.
- Developed template functions translated into drone motion commands and applied ROS for trajectory planning and multi-robot control.
Jan - Jun 2024
Software Engineering Intern
Arista Networks · Vancouver, BC
- Used TACC, C++, and Python to work on OS optimization for high-performance network switches.
- Wrote hardware-independent library code to monitor job completion times and generated histograms for detailed performance analysis.
May 2023 - Jan 2024
Research Assistant, NLP Group
Alberta Machine Intelligence Institute · Edmonton, AB
- Proposed and experimented with solutions in knowledge distillation for autoregressive transformers.
- Used Python, PyTorch, NLTK, and Hugging Face across model development and experimentation workflows.
May 2023 - Jan 2024
Machine Learning Engineer
NRC CAN & CCG, University of Alberta · Edmonton, AB
- Collaborated on a LiDAR transformer model for accurate tree species identification.
- Targeted real-time species identification to assess wildfire vulnerability and support firefighting strategy development.
May - Aug 2022
Research Assistant, High-Performance Computing Group
University of Alberta · Edmonton, AB
- Collaborated with IBM to design and develop a unified MLIR dialect for matrix multiplication on the IBM Power10 MMA unit.
- Applied C++ and Tablegen to build production-quality compiler components.
Toolkit
C++PythonPyTorchROSBazelMLIRTablegenTensorFlowHugging FaceJavaScriptSQLDocker
04 / more projects
More from the workbench.
Small products, research prototypes, and tools built to be used.
01Meal Planner App
A meal planning workflow that turns recipes and weekly plans into an automated shopping list.
View on GitHub ↗
02LaTeX Text-to-Speech
A parser and speech pipeline that turns mathematical notation into natural language and audio.
Open the project ↗
03Imposter Detector
A playful speech-to-text experiment that surfaces suspicious trigger words in real time.
Open the project ↗
04Movie Database Terminal App
A command-line data processing tool for searching, updating, and managing a movie database.
View the code ↗05 / research
Research & publications
My research lives where learning systems meet constraints: safe motion, continual adaptation, and tools that make experimentation easier.
28citations
3h-index
4works listed
2025 · IEEE Robotics and Automation Letters20 citations
SwarmGPT: Combining large language models with safe motion planning for drone swarm choreography
Language-based choreography augmented with a safety filter that makes minimal corrections when feasibility constraints are violated.
Read the paper ↗2024 · arXiv5 citations
SwarmGPT-Primitive: A language-driven choreographer for drone swarms using safe motion primitive composition
Composing deployable drone choreographies from language and a library of safe motion primitives.
Read the preprint ↗2025 · arXiv3 citations
The Cell Must Go On: Agar.io for Continual Reinforcement Learning
AgarCL, a non-episodic, high-dimensional platform for studying agents that learn in changing environments.
Read the preprint ↗2024 · NeurIPS Datasets & Benchmarkspublication
Using Unity to Help Solve Reinforcement Learning
The United Unity Universe, an open-source toolkit for creating varied reinforcement learning environments in Unity.
Read the paper ↗Research thread
Language → planning → motion
My current line of work explores how expressive, high-level intent can become reliable behaviour in multi-agent systems.
06 / field notes
Notes from the workbench
What a compiler IR teaches us about ML systems
Abstractions work best when they preserve the information the next layer actually needs.
Read note ↗From language to flight
Designing a bridge between expressive instructions and the safety constraints of a drone swarm.
Read note ↗A portfolio should explain how you think
A small argument for showing the questions, constraints, and tradeoffs behind the work.
Read note ↗07 / contact
Have a hard problem or a good question?
I am always open to interesting software, research, and collaboration conversations.
vedudx@gmail.com ↗