Digital Paperwork / Fintech / HR Information System / Workforce ManagementTech stack: Typescript, React, React Native, Golang, PostgreSQL, Terraform, Kubernetes, Istio, AWS (EKS, RDS, S3, SQS, SNS, CloudWatch Metrics/Alarms/Logs, API Gateway, Lambda, KMS, Secrets Manager, Step Functions, DynamoDB, EC2, VPC, ELB, ECR), Cloudflare, Sentry, Pagerduty.

After our previous General Manager left the company, I took over the role and narrowed company focus to capture both sides of our marketplace (consumers and businesses) and build brand awareness, shelving paid features for the future. This included setting goals with our board of investors, specifying/collecting metrics, regular reporting, and leading/growing the entire team to hit those goals.In a two year period, we grew from ~200 to ~10k daily unique, organic visits and ~$1k to ~$600k SEMRush value using 100% white hat tactics for growth. We built positive, two-way relationships with over 50,000 businesses.With a constrained budget, I worked with seven direct reports across four departments (engineering, content, marketing, data) to operationalize and offshore as many of our processes as possible, including: data collection/QA, image editing, writing, editing, and marketing. Our in-house team was small, but effectively scaled to manage ~250 contractors on a daily basis across multiple departments.Additionally, I was responsible for all budgeting, payroll, HR, and legal concerns at the company. This included financial forecasts/budgeting, data analyses (Tableau/Excel/Jupyter), and updating HR/legal methods/docs as needed (reworked hiring loop; rewrote hiring docs and employee manual; formalized review/feedback processes).Engineering responsibilities and leadership continued, including: introducing code splitting (via Webpack), migrating from Flux to Redux, releasing a React Native app, and ongoing tool development to scale all business departments. I containerized our services for faster, regular deployment (Docker) and shifted to blue-green deployments (leveraging HAProxy) enabling 99.999% uptime and canary releases when needed. Automated data/writing QA with spaCy. Developed an SVM classifier to identify scammy businesses, leveraging pre-trained GloVe vectors. Migrated marketing/crm stack to DynamoDB, SQS, and Lambda for significant cost savings.

When I started Expertise, I focused primarily on engineering/product needs with a goal of building a fully functional bidding system for a two-sided marketplace (lawyers/clients, contractors/homeowners, etc). I did the initial customer research (businesses/consumers) to create the initial product prototype, with further research and product ownership later handed off to our product team.I built and led a team of six (five engineers and one UX/UI designer). Defined tech stack, owned architecture of the system, managed development lifecycle, grew team members (code reviews, 1-on-1's, professional development goals, etc), participated as a hands-on contributor alongside my teammembers, and ensured the product was built with a long-term vision.Tech stack: an isomorphic ReactJS/Flux frontend (server-side rendering in NodeJS for speed and SEO purposes), Python backend REST API (Pyramid, SQLAlchemy, Postgres), Webpack+Babel for bundling and ES6/ES7 support. Front end and back end designed as independently, horizontally scalable services (back end composed of three microservices). Event-driven architecture on Kinesis + Lambda.Local testing with Vagrant. Deployment to AWS via Chef and Terraform (EC2, CloudFront, ELB, S3, SQS, SNS, Kinesis, Lambda).Additionally, created a number of internal tools to support data collection/QA efforts and marketing purposes, coded in a mix of Python and NodeJS (depending on which language best met each tool's requirements). ZeroMQ for local IPC, REST APIs behind SQS and SNS for queuing RPCs.I prototyped native iOS/Android apps after trying a cross platform Cordova solution which didn't fit our needs well. After seeing the promise of React Native develop rapidly in early/mid 2015, we put app development on hold and revisited it in 2016 after React Native was more mature and proven.

Designed and built fault-tolerant, highly available hosting system with 80ms p95 and 150ms p99 response time. Deployment/management with Chef on AWS EC2: Apache 2.4; caching via Varnish; MySQL with streaming replication and auto failover; automated backups to S3; Logstash, Kibana, Munin, and Graphite for server logging/analysis; Sentry for code error collection/logging; HAProxy and ELB for high availability; Linux Containers (LXC) for site isolation.Performed a broad security audit and updated accordingly at server, platform, and code level.Collaborated directly with designers and copywriters to update old sites and bring them up to spec for usability, browser compatibility, and SEO best practices.(This was an interim position as the company downsized and prepared a portfolio of 200+ websites for sale. Following that, the company pivoted to being a startup incubator/studio.)

Delivered an end-to-end Natural Language Generation ("NLG") system capable of generating millions of pages of readable, coherent, instructive content at a high-school reading level with 3,000 - 5,000 words. Results indistinguishable from an accomplished copywriter, with high variation between pages while capturing complex ideas including analysis of numbers and trends.I began as the team lead and lead engineer and in six months was promoted to engineering manager. My role included: defining the technology stack and project direction, high-level architecture, building/leading a diverse team of twelve, interfacing with company management (collecting specifications, demoing progress, delivering final content), and driving progress to hit goals while contributing hands-on with my team.Since the path to the solution was not well-defined or already solved, I opted to use Scrum with an emphasis on rapidly developing working prototypes, then iterating on those prototypes each step of the way. With a large team and complicated problem, I placed a strong emphasis on code reviews and test-driven development (TDD), maintaining coverage at 95%+ paired with CI (continuous integration, Buildbot with GitHub integration automating quality/standards/test-suite checks). Documentation was regularly updated (via Sphinx).At the lowest level, we built off Michael Elhadad's SURGE grammar (extended as required, run on a simplified LISP-like interpreter coded in Python). On top of that we built a stack of abstractions for phrases, clauses, texts, sentences, messages, and documents which could be manipulated for the purposes of aggregation, references, and variation.Inputs were primarily well-structured data with some sentence structures, entities, and relationships derived from existing text via CoreNLP, combined with custom machine learning models.Large-scale generation on AWS spot instances via MapReduce.

Created new internal tools for data visualization and marketing. Backends developed as SOA to share common services where possible. Java, Python, Pylons / Pyramid, SQLAlchemy, AWS, PostgreSQL with PostGIS, Redis, ZeroMQ.Created a horizontally scalable, highly available, fault-tolerant key-value store for massively parallel data storage and processing. Routing via custom client-side libraries. Distributed processing originally via custom process pools backed by RabbitMQ, later migrated to Hadoop for superior fault tolerance.Drove standards adoption across the engineering team, getting everyone to adopt code style guides, effective Git use, and use off the shelf solutions for security (SHA512 signatures, bcrypt password storage, etc). Introduced SLAs, metrics collection (Munin), and automated testing (Selenium, etc).The company grew from about twenty people to 200 in six months, so occasionally I wore a sysadmin/IT hat. This included: setting up wireless (Cisco) and wired networks (Dell PowerConnect managed switches) with departments isolated on different VLANs, a VPN, and load balanced external routing (WatchGuard / Cisco).

Full-service consultancy, tooling, and web dev shop. Specializing in Python, backends/tools, REST APIs, streaming technologies, fault-tolerant systems, and BI/analytics.Developed an end-to-end solution for a client who needed a high volume, real-time streaming solution in the browser. The system communicated with the browser via HTTP long-polling, backed by a custom, event-driven HTTP server coded in C (with libevent) with Python embedded in a process pool. Python code was treated as synchronous, and C code was treated as asynchronous which enabled their in house devs to easily modify the Python code as needed.Built a fault-tolerant, highly-available in-browser instant messaging platform in Erlang, enabling browser-based customer support for the client's e-commerce business. Concurrency and leader election via Multi-Paxos.Updated a client's webapp for their reservation system to expose a RESTful API and built/published a native, Objective-C iOS app which interfaced with the API enabling customers to book reservations from the app.

GigaOM was one of the first major commercial users of Wordpress. I created and maintained multiple WordPress plugins to enable effective writer/editor/publisher collaboration (workflow tools).Replaced Wordpress's default search page with a vastly improved search page powered by Solr, indexing content from all network sites and creating facets.Additionally, created boutique sites for events/conferences including: user registration, marketing campaigns, and regular site updates. Converted user registration into a standalone service enabling faster development of new sites and reducing administration/operation burden. Prototyped an iOS check-in app.Other tools:- An embeddable polling widget with a “vanilla” JS frontend and a Python backend service (Pylons, SQLAlchemy).- A real time stats collection/processing service (OLAP) to identify early problems or successes in fresh content (rapidly prototyped in Python + Twisted, production in Java for speed+scale).- Event push service, supporting SMS and a web-based polling widget. Used during conferences, supporting rapid 10k+ fanout for events.