Brandon J. Peck

Computer vision & applied machine learning
Philadelphia, Pennsylvania
I work on computer vision and applied machine learning, with a focus on understanding human activity. My interests include learning from synthetic data and evaluating how models perform in unfamiliar environments.
My background includes graduate study in artificial intelligence at the University of Pennsylvania, computer vision research at UVA, and applied science at Microsoft. At Mercury Alert AI, I apply computer vision to fall detection and older adult care.
EmailLinkedIn
Research interests
Computer vision / Synthetic data / Generative modeling / Applied machine learning

Selected research

Research projects, course reports, and a patent application across computer vision and language modeling.
Featured study · September 9, 2026

Fall detection and activity recognition

Brandon Peck · Mercury Alert AI
Compared Mercury Instant with Claude Opus 5 on 15,000 reference-labeled images from ten Mercury camera views across five activity states. A 1,151-image subset supports visual inspection and reproducible comparison.
Model
Fall recall
Fall precision
Macro F1
Mercury Instant
93.53%
98.35%
89.89%
Claude Opus 5
70.17%
98.73%
79.67%
Exploratory image-level evaluation: camera and checkpoint selection involved prior results. Correlated frames limit independence; these metrics do not measure independent fall incidents or deployment alert precision.
Read the paper ↗Explore evaluation ↗Demo ↗
Patent application
Mercury Alert, Inc.

Systems and Methods for Detecting a Position of a Subject

Patent application describing subject position detection using computer vision and neural networks, developed for Mercury Alert's fall detection system for older adult care.
Computer vision · Neural networks · Fall detection
Patent application ↗Demo ↗
Graduate course project
UVA VisLang Lab · Vicente Ordóñez Román

Human Action Recognition with Synthetic Data

Generated 12,000 labeled action videos with Unity ML-Agents and trained an action-recognition pipeline combining a frozen ResNet18 backbone, pose estimation, and an LSTM. Achieved 85% accuracy on 300 real-world videos; selected as a top graduate computer-vision project.
Synthetic data · Unity ML-Agents · ResNet18 · Pose estimation · LSTM
Project report ↗
Course project
NLP Course at UVA

Text Generation with a Sliced Wasserstein Autoencoder

Co-developed a Wasserstein autoencoder for text generation using MS-COCO captions. Reduced test reconstruction error by 6.1% versus a VAE baseline while resisting latent-space collapse.
Wasserstein autoencoders · Sliced Wasserstein distance · MS-COCO captions
Project report ↗Explanatory article ↗
Research assistantship
UVA VisLang Lab · Vicente Ordóñez Román

Feedback Prop for Multitask Networks

Evaluated feedback propagation at inference time in shared-trunk multitask CNNs. Tested whether partial task labels could refine learned representations and improve predictions for unknown tasks without retraining.
Computer vision · Multitask learning · Feedback Prop
Independent study
UVA Link Lab · Madhur Behl

Smart Waste Sorting with Object Detection

Built a waste-sorting system with YOLOv3, CUDA/OpenCV, and Arduino/NeoPixel hardware. Detected four object classes and illuminated the appropriate bin; benchmarked performance across four distances and three camera heights.
YOLOv3 · CUDA · OpenCV · Arduino · NeoPixel
Project report ↗

Academic & professional background

Research affiliation · Researcher
WAVES Lab · University of Pennsylvania ↗
Wireless, Acoustic, Vision, and Embodied Systems Laboratory, led by Prof. Mingmin Zhao. The group studies sensing, perception, and embodied systems using wireless, acoustic, and visual signals.
Education
University of Pennsylvania
M.S.E. Artificial Intelligence
Expected May 2027
GPA: 3.9 / 4.0
Ranked 5th of 158 AI master’s students in an NLP model training competition.
University of Virginia
B.S. Computer Science with Distinction
2015 - 2019
Computer Vision Research under Vicente Ordóñez Román
Microsoft Senior Student Ambassador; founded Microsoft at UVA.
Experience
Mercury Alert AI
Co-founder & CEO · ML Engineering Lead
May 2022 - Present · San Diego, CA
Co-founded Mercury Alert AI and led development of its camera-based safety monitoring system, from PyTorch model training and AWS deployment to caregiver applications.
Lead a seven-person team spanning machine learning, backend, mobile, marketing, and hardware, with responsibility for system architecture and engineering delivery.
Raised venture funding from Wharton Alumni Angels and 2468 Ventures, and co-authored a geriatrics / ML grant proposal and clinical pilot plan with UC San Diego.
Incubated in UCSD MedTech 2 accelerator.
Microsoft
Data & Applied Scientist II
Aug 2019 - May 2022 · Redmond, WA
Designed Surface user segmentation from app-usage telemetry, normalizing activity volume and comparing K-means and DBSCAN for cluster separation, stability, and interpretability.
Productionized persona clustering in Python, PySpark, and Azure Databricks across 10M+ devices and billions of telemetry events; segments became filters across Surface reporting.
Established a SQL telemetry-modeling standard across Surface Data Science and owned the flagship prerelease abnormal-shutdown report used for reliability analysis and launch readiness.
Microsoft
Data Scientist Intern
May 2018 - Aug 2018 · Redmond, WA
Built and deployed a triple exponential smoothing anomaly detector for prerelease Surface telemetry, identifying regressions 1–2 days earlier than prior monitoring.
Validated the system on nine historical datasets; it remained in use after the internship.
DecisionQ
Software Development Intern
May 2017 - Aug 2017 · Arlington, VA
Implemented a K-D Tree and custom nearest-neighbor search in C++11.
Built a C# tool that mapped Java code into Microsoft Visio data-flow diagrams to help engineers understand legacy systems.
Contributed Beautiful Soup scrapers for Deep Web forum data.
Selected recognition
2023
AI Panelist — Medical Alert Monitoring Association ↗
2017
First Place — UVA Global Health Case Competition ↗
2015
First Place & Only HS Finalist — Microsoft Imagine Cup National Competition ↗

Applied work

Putting machine learning into practice through deployed systems and interactive applications.
Computer vision · Healthcare

Mercury Alert AI

Co-founder & CEO · ML Engineering Lead · May 2022 - Present
Mercury Alert AI uses computer vision to detect falls and monitor older adults’ activity through connected cameras, giving caregivers alerts and nighttime activity insights without wearables. I co-founded the company and built the vision models, cloud inference pipeline, camera integrations, and caregiver app.
•
Designed and optimized a PyTorch fall-detection and activity-monitoring model across 2,800+ training runs and 460+ configurations, with evaluation on unseen device environments.
•
Scaled production inference with SageMaker, Lambda, and S3 to support 1.6M+ frame evaluations per day and a 73M-object, 12.4 TB image corpus.
•
Integrated Ring cameras through an event-driven AWS pipeline and shipped a React Native / Expo app with PostgreSQL-backed alerts and nighttime activity insights.
PyTorch · SageMaker · AWS Lambda · Amazon S3 · React Native · Expo · PostgreSQL
Website ↗App Store ↗Amazon product ↗Ring Appstore ↗
Applied AI · Social systems

Circles Verified

Creator · 2023 - Present
Built a social application with AI-driven affiliation verification, connecting people through shared communities and backgrounds.
React Native · AI/ML · Supabase · TypeScript
App Store ↗Demo ↗
Generative AI · Interactive narratives

VirtuFriends

Creator · 2024 - Present
Built an interactive storytelling application using generative AI for character-based narratives and conversations.
React Native · Generative AI · Hugging Face · Supabase
Website ↗App Store ↗

Methods & tools

Methods
Vision transformers · Transfer & self-supervised learning · Sequence modeling · LLM pretraining · BERT fine-tuning · Retrieval-augmented generation
Modeling & analysis
PyTorch · Torchvision · timm · scikit-learn · TensorRT · NumPy · pandas · SQL
Languages
Python · JavaScript · TypeScript · Java · C++ · C
Frameworks
Hugging Face Transformers / Datasets · SentenceTransformers · OpenCV · React Native · Expo
Infrastructure
SageMaker · AWS Lambda · Amazon S3 · PostgreSQL · DynamoDB · SQS / EventBridge · Azure Databricks · Git · Docker · Stripe

Contact

I'm interested in conversations about research, applied machine learning, and opportunities to collaborate.
brpeck@engineering.upenn.eduLinkedIn ↗
© 2026 Brandon J. Peck