Server Benchmarking for vLLM, Luminal and SGlang Inference
I got the opportunity recently to work on Luminal(YC S25)'s inference server benchmarking tool. I'm using Prometheus and coroutine based algorithms to get the most accurate server metrics for every LLM delivering high-fidelity latency/throughput metrics (p50–p99.9), CPU/GPU utilization, and per-model comparisons across backends.
DockSleek
DockSleek is a Go-powered tool that analyzes and rewrites Dockerfiles to reduce image size, speed up builds, and improve caching. It suggests multi-stage builds, collapses redundant layers, and strips out unnecessary artifacts, delivering leaner, faster, and more secure containers.
This is becoming important as I get deeper into DevOps.
Currently a WIP
ContextQ
Fit your LLM to any task with 1 line! Saves up to 80% of size and cuts down inference by 2x minimum.(AWQ) Context-Aware Model Library for task specific LLM quantization.
pip install contextq
Chord Wizard
ChordWizard is a real-time chord recognition system using OpenCV and spectral analysis. Processes audio input through Librosa for MEL spectrogram generation, then uses CNNs to detect and identify guitar chords. Users can create their own progressions and there is also a built in finger position visualizer.
Server Benchmarking for vLLM, Luminal and SGlang Inference
I got the opportunity recently to work on Luminal(YC S25)'s inference server benchmarking tool. I'm using Prometheus and coroutine based algorithms to get the most accurate server metrics for every LLM delivering high-fidelity latency/throughput metrics (p50–p99.9), CPU/GPU utilization, and per-model comparisons across backends.
DockSleek
DockSleek is a Go-powered tool that analyzes and rewrites Dockerfiles to reduce image size, speed up builds, and improve caching. It suggests multi-stage builds, collapses redundant layers, and strips out unnecessary artifacts, delivering leaner, faster, and more secure containers.
This is becoming important as I get deeper into DevOps.
Currently a WIP
ContextQ
Fit your LLM to any task with 1 line! Saves up to 80% of size and cuts down inference by 2x minimum.(AWQ) Context-Aware Model Library for task specific LLM quantization.
pip install contextq
Chord Wizard
ChordWizard is a real-time chord recognition system using OpenCV and spectral analysis. Processes audio input through Librosa for MEL spectrogram generation, then uses CNNs to detect and identify guitar chords. Users can create their own progressions and there is also a built in finger position visualizer.
Ayan Jhunjhunwala
Focused on full-stack development, distributed systems, and applied LLM research
Computer Science @ UC Irvine
Focused on full-stack development, applied LLMs + data driven systems and devops!
Computer Science @ UC Irvine
Server Benchmarking for vLLM, Luminal and SGlang Inference
I got the opportunity recently to work on Luminal(YC S25)'s inference server benchmarking tool. I'm using Prometheus and coroutine based algorithms to get the most accurate server metrics for every LLM delivering high-fidelity latency/throughput metrics (p50–p99.9), CPU/GPU utilization, and per-model comparisons across backends.
DockSleek
DockSleek is a Go-powered tool that analyzes and rewrites Dockerfiles to reduce image size, speed up builds, and improve caching. It suggests multi-stage builds, collapses redundant layers, and strips out unnecessary artifacts, delivering leaner, faster, and more secure containers.
This is becoming important as I get deeper into DevOps.
Currently a WIP
ContextQ
Fit your LLM to any task with 1 line! Saves up to 80% of size and cuts down inference by 2x minimum.(AWQ) Context-Aware Model Library for task specific LLM quantization.
pip install contextq
Chord Wizard
ChordWizard is a real-time chord recognition system using OpenCV and spectral analysis. Processes audio input through Librosa for MEL spectrogram generation, then uses CNNs to detect and identify guitar chords. Users can create their own progressions and there is also a built in finger position visualizer.
Skills
Skills
Python
Python
C++
C++
.NET
.NET
TypeScript
TypeScript
React/Angular
React/Angular
Node.js
Node.js
Full Stack
Full Stack
SQL
SQL
Pytorch
Pytorch
Golang
Golang
Huggingface
Huggingface
Terraform
Terraform
Docker
Docker
Prometheus
Prometheus
Git
Git
Apache Spark+Airflow
Apache Spark+Airflow
AWS Stack
AWS Stack
Azure
Azure
ML/AI
ML/AI
Experience
Experience
AI Engineering Intern (Summer) + SWE Intern (Fall)
AI Engineering Intern (Summer) + SWE Intern (Fall)
LBA
LBA
June 2025 - Present
June 2025 - Present
Machine Learning Researcher
Machine Learning Researcher
UCI Health
UCI Health
March 2025 - Present
March 2025 - Present
Accelerate Fellow
Accelerate Fellow
IBM
IBM
May 2024 - July 2024
May 2024 - July 2024