Joss
Steward
| Location | Seattle, WA |
| joss@42hz.net |
Elixir Software Engineer
SUMMARY
Skilled software engineer and architect with a proven track record in developing desktop, cloud and embedded applications. Experienced with many platforms, and always interested in learning new technologies and languages. Consistently recognized for high quality work. Large breadth of knowledge, with particular interest in machine learning, scalable systems, and video analysis and delivery. Very fast learner and efficient self-directed worker.
TECHNICAL HIGHLIGHTS
Proficient:
-
Elixir, Erlang & OTP, Golang, EXLA, NX, PostgreSQL, CUDA, ECTO
-
Applied LLMs, Multimedia processing and analysis
-
System Architecture, Microservices, Agile & Test-Driven Development
-
CloudFormation, Terraform, CDK, Docker, Hex Package Distribution, Debian Packaging
-
AWS Cloud Application Components including DynamoDB, SQS, Lambda, Fargate, EC2, Route 53, CloudFront, Elastic Load Balancers, RDS, and S3 as well as equivalent products from Azure and other providers.
Professional Experience with:
-
Java, Python, Lua, Golang, C, C++
-
Pytorch, Agentic System Design, Clustered System Design and Administration
-
HTML, CSS, JavaScript
-
FreeBSD and Linux Server Administration
-
PostgreSQL Server Administration, ElasticSearch, Kibana, Grafana
EXPERIENCE
2022 - 2026
Wurl
RemoteStaff Elixir Software Engineer
Staff Elixir Software Engineer at Wurl filling lead developer and architect roles on several projects, reporting directly to senior leadership. Projects included modernization efforts on legacy services as well as greenfield development of a new LLM-powered video analysis and segmentation service. Gained a lot of experience working with the specifics of HLS video distribution and common industry tools including ffmpeg/ffprobe, media tailor, pytorch, whisper, and various content distribution networks. Primarily worked with Elixir, Go, and Python.
-
Lead on the design and implementation of an agentic video segmentation service which leveraged tool-calling and other structured LLM interactions to categorize, summarize, and automatically select interesting and compelling segments for promotion from 500k+ hours of archived content. The service we designed and built ultimately provided a full-fledged video analysis toolchain including a structured job system with dependencies and caching for various forms of video metadata, automated transcription using whisper and Bumblebee, scene detection, asset segmentation, A-B testing for prompt development, and experimental features.
-
Our agentic video segmentation service was built in a combination of Elixir and Python, and deployed onto EC2. The design took full advantage of several OTP features, using ETS and process linking to build a dynamic asset cache for managing large video assets, LiveView for the administrative interface. We used NX for much of the machine learning when possible, and fell back to python when we needed libraries and tools that weren't available in Bumblebee yet. Jobs were orchestrated with Oban, and as of today this system has successfully processed > 1 million hours of video and > 4 million separate Oban jobs. All the metadata from analyzing video was stored in PostgreSQL and QDrant.
-
Lead a team effort to replace a legacy system responsible for electronic program guide (EPG) delivery to various content partners. With a short turnaround time we successfully designed and delivered a complete solution to replace the existing ad-hoc system with a cheap and low-maintenance service that was deterministic and driven directly by a database with an API for integration with other internal systems.
-
Implemented several state-of-the-art video analysis techniques as described in whitepapers and industry literature, and rapidly delivered several of the described algorithms from prototype to production use. Projects included visual embedding powered scene boundary detection, automated time-aligned transcription, and more.
-
Designed, tested, built, and deployed a simple test-tone detection algorithm using NX. This worked by stripping the audio track out of a source video and running it through a discrete fourier transform before taking the cosine similarity between adjacent frames of the FFT output. This resulted in a very clear signal where the DTMF test tone was immediately identifiable with an accuracy exceeding 99.99%. Total time from feature request to deployed implementation was 2 weeks.
-
Built a suite of video analysis and manipulation tools in LiveBook by embedding an open source video player - amalia.js - and using the existing infrastructure from our agentic video segmentation service to provide a complete toolset for developing novel video analysis algorithms. This greatly simplified the process of tuning transcription alignment, scene detection, and similar work by making it easy to test algorithms against real data without rerunning long analysis tasks each time.
-
Built and demonstrated a prototype HLS playout system in Elixir using OTP tools including GenServers, a custom registry, dynamic supervisors, and ETS which managed live streaming playout of HLS from VOD assets. This had a number of advantages over the existing playout approach, and included the ability to dynamically change and reload the playout schedule without interrupting playback. Part of this included implementing the HLS RFC8216 in Elixir, using NimbleParsec for the parsing and some custom macros for serialization.
2021 - 2022
Tanium
RemoteSenior Software Engineer
Member of small squad responsible for maintenance and bug-fixing on Tanium Plugin Modules. Responsibilities include verifying bug reports, writing automated tests to reproduce issues, improving the quality of older codebases, and developing patches. I also mentored new team members and worked to improve the reliability and ergonomics of our internal tooling.
-
Used LiveBook extensively to parse and debug network protocol issues.
-
Took a course on server-side Typescript.
2019 - 2021
Amazon.com Web Services
Seattle, WASoftware Development Engineer II
Member of AWS Corporate Collaboration team, responsible for providing email services for Amazon including AWS and subsidiaries. Maintained, improved, and developed new services responsible for anti-spam and security, mail transfer and delivery, mailing list functionality, mailing list moderation tooling, and monitoring infrastructure. The globally distributed mail system handles an increasing load currently at more than 1,500 deliveries per second.
-
Led design and implementation of Attachment Control System for outbound emails, designed to enable the sender to check if attachments have been viewed and allow the user to revoke access from unintended recipients.
-
Designed and built a next-generation email address governance system that leveraged DynamoDB to improve replication speed, exibility, and availability of the email address and list management system. Added custom LDAP shim to interface with postfix and replace a system of flat files.
-
Contributed to the design and implementation of a replacement email list moderation system, vastly improving the experience provided by the existing moderation tooling and adding new features, including the automation of some policy-driven moderation decisions.
-
Designed and demonstrated prototype next-generation mail transit infrastructure, replacing traditional postfix systems with a cloud-native solution leveraging S3, DynamoDB, and Lambda to improve delivery reliability, throughput, and configurability.
-
Build and demonstrated SMTP parsing logic in Elixir using NimbleParsec, and used the parser to build a simple SMTP server. This was experimental, but the IPC capabilities of Erlang proved a nice match for the requirments of an SMTP server.
2017 - 2019
Saft America - Space and Defense
Cockeysville, MDEmbedded Software Engineer
Embedded Software Engineer working on control systems for Advanced Lithium-Ion Batteries in the Saft America Space & Defense Division. Responsible for developing and supporting safety-critical, bare metal, hard-real-time embedded monitoring and control programs for high voltage energy storage systems using C and C++, testing products and changes to validate specifications and troubleshooting failures in returned products.
-
Received "Star" award for work done to test and deliver an important product to the customer on time in December 2017.
-
Designed and implemented a custom UDP-based monitoring protocol optimized for low power consumption and low overhead on resource-constrained microcontrollers.
-
Developed chassis-mounted HMI using embedded Linux and QT.
-
Developed and implemented control algorithm for enhanced high voltage bus activation after completing analysis and simulations, increasing safety and fault tolerance of prototype kilovolt energy storage system with peak discharge power exceeding several megawatts.
-
Updated existing and created new desktop applications used for debugging, testing, and monitoring products.
-
Improved and expanded automated hardware testing platform.
2016 - 2017
First Data Corporation
Hagerstown, MDSoftware Engineer; Intern
As one member of a small team I was responsible for developing and supporting line-of-business applications using Java and SQL. Our largest project was replacing an existing book-balancing solution which relied on Microsoft Excel 98 and several thousand lines of visual basic macros with a more reliable and debuggable solution written in Java.
-
Became familiar with Microsoft Visual Basic and Oracle SQL RPC.
-
Our project to replace the legacy system was succesful, and removed the ongoing need to maintain windows XP virtual machines just for line-of-business work. The replacement followed industry best practices and had extensive logging, emitted appropriate metrics, and was run by a standard cron system instead of the previously used custom windows service.
2012 - 2015
Volvo Group Trucks
Hagerstown, MDSoftware Engineer; Intern
Worked during summers for over 4 years as part of a team developing software to support research and engineering being performed in the Product Development Laboratory.
-
Developed C# data analysis and reduction program for emissions compliance testing, eliminating 2 weeks of manual analysis work and allowing emissions compliance teams to remain in the field uninterrupted.
-
Sole developer responsible for the creation of an ASP.Net website and associated windows data gathering service used to monitor and analyze utilization of heavy-duty diesel engine testing equipment over time.
2012 - 2013
Collaboratory
Grantham, PAProgrammer - Volunteer
Developed embedded firmware in C and Assembly to run on custom hardware. The project goal was the development of a remote monitoring system to detect and warn of impending hardware failures in hand-operated water pumps in rural Africa, reporting pump seal health over SMS.
EDUCATION
2013 - 2017
Bachelor of Science in Software Engineering
Shippensburg UniversityInvolved in Video Game Development club and Competitive Programming Team. Worked with fellow students on many small projects including the construction of a small robotic car for competition in the Freescale Cup, the creation of a student-run server used for penetration-testing practice, and a distributed prime-number finder written in Haskell.
-
Used Erlang on a research project related to river water quality in the Susquehanna River Basin. The goal of the project was to explore various stream-processing technologies. Landed on a design where each watershed sensor was mapped to a single GenServer, which would process the data and emit rolling averages as well as alerts when data deviated significantly from historical norms.