Tom Mitchell Email & Phone Number

Seattle, Wa, Us

New York City, Ny, Us

Seattle, Wa, Us

Recent part-time engagement as sole founding engineer for a web data extraction and competitive insights startup.• Built out Python codebase from the ground up, including web crawler, data models, supporting scripts, documentation, and engineering infrastructure. Designed to be extensible, scalable, and understandable for easy transition to a new technical cofounder.• The fully-functional data product has been launched and sold to multiple B2B customers.• The crawler is an extensible platform built on top of Scrapy, backed by PostgreSQL. With minimal oversight, it regularly evaluates 9000 identifiers across 100k+ websites, using 400 parallel shards. At this level of parallelization, it can complete a crawl in about 6 hours.• Terraform is used to provision all AWS resources, utilizing a highly scalable architecture on top of Batch and Step Functions, writing results to RDS. Bazel is configured as the build system for testing, packaging, and deploying code to AWS Lambda and Docker images.• Low-code internal tooling with Retool supports the configuration and operation of the system by non-technical users without significant engineering investment - allowing exploration of new identifiers to boost sales prospects, and monitoring the status of running jobs.

New York, New York, Us

Returned to a Tech Lead and Manager role for the Ocelot Adjudication team. Overseeing a team of five engineers through continued development of Oscar's next-generation claims processing engine. Responsible for formulating strategic medium- and long-term goals, managing progress and milestones, and providing technical leadership. Business areas covered include authorizations, cost sharing calculation, benefit determination, accumulator ingestion, repricing, adjustments, and many others.

New York, New York, Us

Returned to Seattle WA, working full-time remote as an IC on Oscar's Engineering Infrastructure team.Built an internal tool from scratch which allows users to query any Oscar gRPC service using a dynamically generated React form, according to the corresponding Protobuf request schema. The tool is utilized by engineers and non-technical users alike.Contributed to internal infrastructure designed to move data from a MySQL or Postgres replication log into various sinks, such as BigQuery or Kafka. This centrally-developed and generic tooling replaced multiple bespoke solutions which had cropped up independently across the organization.

New York, New York, Us

Took over management responsibilities for the Plans and Benefits team.

New York, New York, Us

Bootstrapped a brand new backend team responsible for managing Plans and Benefits data across Oscar; including data models, services, configuration tools, benefit adjudication business logic. Led the team through growth up to 5 new engineers.

New York, New York, Us

Founding member of a new team spun up to start work on Oscar's inhouse claim processing engine. Involved in core early decisions around the technology stack, such as the service architecture, data storage scheme, and domain-specific language for developing business logic. Designed and implemented several core features of the system such as accumulators and generic object lookup storage. Created initial placeholders for components such as benefit determination, cost share calculation, member reconciliation.Designed and implemented a storage platform for bitemporal data, on top of PostgreSQL. Bitemporal data includes two time dimensions for every row - real-world time to represent how an entity changes over time, and database time to represent how our understanding changed over time (historical record). The platform includes codegen tools to produce SQL definitions, ORM definitions, service definitions, etc from a single datasource YML definition. Entity history is managed automatically, and database constraints prevent bitemporal inconsistency. Libraries were developed for Python and Go specifically to manipulate bitemporal data. The platform is being utilized by multiple teams at Oscar.Designed and implemented an ingestion service for reconciling raw claim member data (often with data entry errors) against our internal member database, using Elasticsearch for fuzzy matching. Optimized query parameters to minimize false positives.Designed and implemented a data staging area for an external data ETL pipeline, allowing full pipeline runs to complete and validate results, without impacting live data if something goes wrong. Supporting infrastructure included jobs for setup/teardown of the staging area, and validation of new vs old data.

Redmond, Washington, Us

Developer on Cortana's Proactive Ranking team.Designed and implemented a pipeline for making pre-computed ranking features available and updated on a daily basis in the online stack. This pipeline can be leveraged by data scientists to quickly onboard new features, with a minimal overhead, and no required knowledge of the intricacies of Cortana's online infrastructure.Designed and implemented a ranker capability called Stacking - allowing proactive cards to be selectively re-arranged underneath other cards. This is a generalized feature, but was initially built to enable a scenario where score predictions for a particular team need to be shown under the game schedule card for that same team.Overhauled the ranker infrastructure to improve the failsafe/fallback story, and eliminate an unnecessary development code branch.

Redmond, Washington, Us

Developer for Bing in the Contextual Relevance group.Migrated several large Contextual Relevance features and components into a new backend framework (internal to Bing). Old code in C++ needed to be translated to C#, with a completely new service paradigm.- Personally responsible for re-writing an existing complex feature (Category-Promoted Search) from scratch - work which took many months to complete and validate, ultimately shipping online code containing fewer discrepancies with offline feature computation.- Assisted another team member with migration of the machine-learned ranker infrastructure powering Contextual Relevance reranking (CRanker).Designed and built a feature called Ranking Stability, which utilized online storage to effectively cache personalized query results per-user, in order to suppress instabilities caused by both the underlying index stack, and the shipping of "cache-busting" in Bing (resulting in a whole new request being sent on a back-button press). This feature was crucial in mitigating this high visibility issue in search result stability.Aided in design and implementation of the User-Annotation Service - a centralized service that exposes a variety of user information into a single easy-to-use schema for partners in the Bing search stack. Migrated some user modeling code to this service - allowing the models to be used by a wider array of customers without code reuse.Strove for high live site quality - continuing to push for improvement to tests and responding to a couple of live site incidents.

Redmond, Washington, Us

Developer for Bing in the Contextual Relevance group.Set up and maintained a small monitoring flight (including alerts) for Contextual Relevance features, which checks daily for excessive changes in coverage or relevance in individual signals. This flight was still running at last check (April 2015). This monitoring was crucial for detecting long-term regressions in the complex machine-learned system, which might otherwise go undetected.Worked closely with folks in MS Research to build out several promising Contextual features in the live site - bridging an expertise gap and providing a means for their research to be put in front of real users. Work was completed primarily in C++.Performed a monthly branch integration (complex source control resolution) for a large code branch - significant amount of work with a lot of dependencies.

Redmond, Washington, Us

Tester for Bing in the Contextual Relevance group.Maintained a variety of tests and followed up on their failures, including unit tests, black box tests, performance tests, and live site monitoring - for a number of different features in the Contextual Relevance search stack (including Suggested Sites, CPS, PNav and CNav). Drove for a high quality bar in terms of reducing false positives and improving test coverage. Notably, no live site incidents were caused by my team during this period.Performed a deep dive on a word-based classifier used to categorize web documents, in order to analyze existing quality and provide recommendations for future classifier improvement.Performed at-scale testing of some critical web index infrastructure in order to assess capacity for exposing additional document features higher up the stack.

Auckland, Nz

Developed software in a team environment on large scale projects, dealing with issues such as version control, requirements conflict resolution and product testing.Part of a team of about 10 developers working on a wide range of projects.Directly involved with software development in several real-world custom built data-oriented applications and websites. Projects included:- A government-sector quota administration system, required to be auditable and keep a history of actions taken. Includes a custom Electronic Data Interchange (EDI) protocol and back end for handling automated quota requests.- An internal business offering for web-based job management, task tracking, billing, and customer records, etc.- A high-volume vehicle dismantling operation in Japan, consisting of both a public website and multiple internal systems handling everything from vehicle purchase, to dismantling, to customer invoicing; all interfacing with a central MSSQL database.Predominantly working with ASP.NET, VB.NET, C#, T-SQL, and Access platforms.Used many supporting tools, including SVN, Git, NUnit, WatiN, CruiseControl.NET, and internal wikis. Personally pushed for the use of Git and NUnit on projects.Produced and worked on a number of frequently-used internal tools, such as a database schema synchronization tool and website scraper.Responsible for introducing testing frameworks to a number of established projects.Appointed as project lead on two ongoing support contracts, communicating with clients daily and overseeing system development. Involved in all stages of a “requirements, development, testing, deployment” rapid software lifecycle, often in direct communication with clients. Required to provide time estimates and client training as part of this process.Responsible for occasional server and network configuration and maintenance, both internally and on-site for clients.