Saab Internships
Overview
Three summers at Saab across three distinct projects — each building on the last. Starting with sensor fusion and data pipelines (2023), moving to computer vision and synthetic data generation for the US Navy (2024), and culminating in a production-grade real-time object detection engine in Unreal Engine (2025).
Summer 2025 — Real-Time Detection Pipeline
Role: Machine Learning Software Engineering Intern · Remote (San Diego, CA)
YOLOv11 Model Training
Trained a YOLOv11 object detection model using 10,000 synthetic images generated under varied lighting, weather, and environmental conditions. The model achieved 89% precision when evaluated against real-world drone datasets — a significant result given it was trained entirely on synthetic data.
Real-Time Async Detection Pipeline
Engineered a fully asynchronous, real-time object detection pipeline inside Unreal Engine using C++. The pipeline integrates the YOLOv11 model via NNE (Neural Network Engine) and OpenCV, processing live camera feeds with minimal latency while maintaining stable frame rates in complex simulation environments.
Custom Messaging Bus
Designed and implemented a custom topic-based messaging bus inside Unreal Engine to route and process real-time sensor data between multiple concurrent systems. This architecture decoupled the detection pipeline from downstream consumers, enabling modular system design and easier integration of new sensors or algorithms.
Summer 2024 — TSUNOMI & Synthetic Data (ONR)
Role: Machine Learning Software Intern · Remote
TSUNOMI was a collaboration with the Office of Naval Research (ONR) to build a multimodal sensor fusion system for the US Navy — combining real-time object tracking, computer vision, and synthetic data generation at scale.
Computer Vision System
Developed a robust camera vision system for tracking objects and providing real-time location data. The system maintained consistent tracking accuracy across varying environmental conditions and object velocities, with an optimized processing pipeline to minimize latency.
YOLO Model Training Pipeline
Designed an automated image pipeline that generates bounding boxes around objects of interest, significantly streamlining YOLO model training. This reduced manual labeling effort and improved the consistency of training data across diverse scenarios.
Unreal Engine Synthetic Data
Implemented a camera calibration pipeline within Unreal Engine to ensure proper alignment between virtual and real-world camera parameters. Created photo-realistic environments — using Cesium for geospatial accuracy — that closely mimicked real-world conditions for AI model training.
Cross-Functional Leadership
Led cross-functional team meetings with university research collaborators, aligning development goals and ensuring innovations could be integrated into broader research initiatives.
Summer 2023 — TSUNOMI, ETA/ETD IRAD & ReflexAI
Role: Machine Learning Software Engineer Intern
Three concurrent projects: a multi-sensor tracking system (TSUNOMI), an aircraft arrival/departure prediction model (ETA/ETD IRAD), and an autonomous boat control system (ReflexAI).
TSUNOMI — Simulation & Kafka Infrastructure
Built a 3D simulated environment using Stone Soup to generate training data for the tracking algorithm. Set up an Apache Kafka broker via Docker to enable seamless, real-time data streaming between multiple sensors and the tracking algorithm. Also developed an acoustic sensor measurement model using covariance matrices and Gaussian distributions to add realistic noise to simulated sensor data.
ETA/ETD IRAD — Aircraft Prediction Model
Engineered a weather data pipeline using the Open-Meteo API to gather historical and forecast weather data along flight paths, storing results in a structured database. Built a TensorFlow machine learning model leveraging this weather data alongside FAA records to predict aircraft arrival and departure times — achieving ~90% prediction accuracy.
ReflexAI — DARPA Demo
Observed and participated in a DARPA demo presentation for ReflexAI, a machine learning model for autonomous control of a boat with dynamic variables (weight, hull integrity, engine power). Gained direct exposure to how defense-focused product demos are structured and evaluated.
Key Technologies
| Category | Tools |
|---|---|
| ML / CV | YOLOv11, YOLO, TensorFlow, OpenCV, NNE |
| Simulation | Unreal Engine, Cesium, Stone Soup, Docker |
| Languages | C++, Python |
| Infrastructure | Apache Kafka, SQL, Git, JIRA |
| Other | ArduPilot, cross-functional leadership |