Ishaan Patel

MS Computer Science at NC State

I'm a software engineer completing my master's at NC State, focusing on machine learning and scalable software systems. I'm seeking full-time backend and applied AI engineering roles.

New grad · May 2027

Oct 2026 – Present

Quazar

Software Engineer

Developing backend services that monitor production systems, prioritize incidents, and automate parts of the development workflow.

Aug 2024 – Jun 2025

Prometheus Group

Software Engineer · C#/.NET, React

Full-stack development in C#/.NET and React, plus the automated testing framework that cut regression time by 40%.

Jun – Aug 2023

Fidelity Investments

Software engineer intern · Java, Spring Boot, Azure

Java Spring Boot services and Azure Service Bus pipelines for 401(k) transaction processing.

Jan 2023 – Jan 2024

UNC Charlotte

Undergraduate researcher · C++, HPC

Adapted Microsoft's ALEX learned index to dynamic graph workloads in C++; OpenMP parallelization for up to 20x speedup.

Jun – Aug 2022

Fidelity Investments

Software engineer intern · Java, Spring Boot, AWS

Extended Fidelity Bloom's Round Up feature with percentage and custom dollar amounts; built user-preference REST APIs on AWS Lambda.

M.S. Computer Science, NC State · ML/AI concentration · GPA 3.78

B.S. Computer Science, UNC Charlotte · Software Engineering · Magna Cum Laude · GPA 3.85

01

Eir

Clearer handoffs between ambulance crews and the ER

I led a five-person team building Eir for the SAS Hackathon 2026, developing core app features, connecting it to SAS Viya, and creating its benchmark workflow. Eir transcribes ambulance reports, extracts vitals that clinicians can check against the transcript, and suggests an ESI triage level for clinician review.

02

Upwind

Explains why air quality spikes across the US, instead of just displaying an AQI

A live app that pulls from 11 sources, including AirNow monitors, weather, NASA FIRMS fire hotspots and satellite aerosol data, to rank the likely causes of a local air quality spike. An LLM judge checks each explanation against the data and replaces failed ones with a rule-based fallback. Nightly evals fail the build if quality drops.

03

crack-seg

Finds cracks in bridge concrete from drone photos

A custom U-Net with sliding-window inference over high-resolution UAV imagery, so thin cracks stay visible without blowing GPU memory. Overlapping tiles are blended with a Gaussian weight map; the current checkpoint adds squeeze-and-excitation and deep supervision.

· · ·

Raw UAV bridge concrete input
Detected cracks overlaid on the input

Drag to compare

04

Pathfinder

Model tuning with help from your IDE agent

A hyperparameter optimization supervisor that uses Optuna to select parameters and prune underperforming trials. It gives your IDE agent the experiment setup, results, and failure history, so you can ask it to compare runs, investigate failures, and help decide what to try next while training continues.

05

BuoyNet

Model compression tradeoff study for solar-powered ocean buoys

An edge deep-compression study for classifying marine microplastics on solar-powered IoT buoy cameras: quantization and pruning under simulated Raspberry Pi 4B constraints, measured against a LoRaWAN payload budget.

Compression study32-bit
Accuracy (%) BuoyNet model size versus accuracy Five measured variants range from 5.98 MB at 98.91 percent accuracy to 1.52 MB at 91.60 percent accuracy. The slider selects a variant. 1009590 06 Model size (MB)
5.98 MB98.91% accuracy

Notes on Learned Indexes

What I learned from a year of research on ALEX and other learned index structures.

Read the notes →