Specialize in managing remote teams, providing constructive code reviews, leveling up junior developers through mentoring, and balancing quality and performance with stakeholder needs and deadlines.
Have architected, developed, and shipped software products for web, desktop, and mobile professionally for a decade.
I was a tpyical webmaster/script-kiddie editing live sites over FTP in the late 90s and early 00s before devops was a concept. Along the way I wrote a number of frontend and backend tools and frameworks and explored dozens of languages, with lasting affection for Ruby and Crystal.
Was involved in numerous DoD and DoE-funded research projects involving artificial neural networks, computer vision, distributed systems, and static analysis.
- Deep Learning
- Data Structures
- Computer Vision
Leading ground-up rebuild of core infrastructure and services (multi-tenant SMS-based learning system)
Scaling up processes and systems to handle exponential growth and web-scale SMS traffic
Building world-class team of backend and frontend developers
Designing ML-based approach to content-based SMS message deliverability detection
Managed 4-person team of junior and senior devs
Architected web-scale, global, low-latency infrastructure for a privacy oriented search engine
Deployed infrastructure, wrote algorithms, and designed systems to circumvent Bing and Google anti-scraping measures with low latency and at scale
Managed and mentored a team of junior developers across three web and one mobile product
Designed serverless Rails app and frontend that drives 360 degree CRE virtual tour platform
Leveraged artificial neural networks to automatically generate and furnish 3D scenes from floorplans
Major contributor to Rails-based CRM for financial advisors (WealthBox)
Triaged bugs, mitigated bottlenecks, built features
Wrote high performance PostgreSQL queries and indexing schemes
Developed high-speed database anonymizer in C++
Major contributor to Rails-based college application platform used by Harvard, JHU, Princeton, etc.
Mentored junior developers and made architecture decisions across a suite of high-traffic products
Developed an analytics product that provided in-depth multi-tenant traffic analysis for college applications
Created an internal DevOps gem enabling encrypted, event-based command-and-control communication between web servers
deep learning • computer vision • machine learning • automated target recognition • algorithms development
Founded the RAPTOR computer vision project, which trains neural networks on synthetic high quality 3D renderings for object detection and 3D pose estimation
Assisted AI research team in dissecting Pedro Domingos' infamous Sum Product Networks paper
Worked with ROSE compiler team to add static analysis code security "checkers" to Compass
Devised a novel technique for efficient memory leak detection in C/C++ code
Received medal for placing in top 10% at the LLNL 2013 Summer Research Symposium
- Johnson, Samuel, "Fast Type-based Indexing and Querying of Dynamic Hierarchical Data" (2017). Brown University Computer Science Master's Theses. https://cs.brown.edu/research/pubs/theses/masters/2017/kelly.samuel.pdf
- Kelly, Samuel, "Toward Decentralized Code Signing: A Legal Framework for Ensuring Software Integrity" (2017). DuroSoft Technical Reports. Paper 001. https://github.com/DuroSoft/PeerSign/blob/master/paper.pdf
- Kelly, Samuel, Jeff Byers, and David W Aha, "RAPTOR Technical Report" (2014). AIC-15-031. https://www.nrl.navy.mil/itd/aic/content/raptor-technical-report
- Kelly, Samuel, "AST Indexing: A Near-Constant Time Solution to the Get-Descendants-by-Type Problem" (2014). Dickinson College Honors Theses. Paper 147. https://scholar.dickinson.edu/student_honors/147
OPEN SOURCE PROJECTS
A C++ driver I wrote that allows disabled users or those who prefer Nintendo Joycons as an input device to use Joycons as a virtual Xbox controller in Microsoft Windows.
A framework I wrote that allows for the deployment of crystal language code to Google Cloud Functions.
A crystal shard I wrote that provides an ActiveRecord-like adapter to MongoDB.
DSeg comes from a novel feature extractor I devised in grad school for object detection. It is based on the idea that unique superpixel shapes arise for different classes of objects.