Skip to work history

Temesxgn Gebrehiwet

I build the backend systems that products run on.

Senior engineer with twelve years of building and running production systems. I own the work from architecture to release, alongside product, design, customer success and the business.

Status
Open to new roles
Role
Senior Software Engineer
Based in
Dallas–Fort Worth, TX

Work

Nov 2023 – Present

Senior Software Engineer, Platform

Celerium, Washington, DC

Cybersecurity SaaS. I own the ingest path of Hyperdark, its Go platform of 30+ services.

  • Standardized 30 firewall syslog parsers onto one shared pipeline, so adding a vendor is an adapter instead of a fork.
  • Replaced CloudWatch ingestion with a multi-tenant Kinesis consumer (parallel shards, Redis checkpoints) that stores no customer keys.
  • Built a Kafka rollup framework and four production workers on it, turning per-event writes into daily aggregates.
  • Ported billing from Python and Ruby to a Go Lambda on Step Functions, synced with Chargebee and HubSpot.

Go, Python, PowerShell, React, AWS, Azure, Kafka, Redis, Kubernetes, PostgreSQL, Prometheus, Grafana

See the system 5 parts

Firewall logs in, threat data the product can trust out

Pick a part to see what I did on it.

Celerium: Firewall logs in, threat data the product can trust outFive parts, left to right. Cloud integrations: log forwarders and blocklist sync that a customer deploys in Azure or AWS. Ingestion: syslog listeners, a listener for firewall managers, and Kinesis consumers on parallel shards with Redis checkpoints. Parsing: 30 vendor parsers on one shared pipeline, and a catalog that normalizes firewall actions. Rollups: four Kafka workers on a reusable framework. Billing, beside them: ported to Go on Step Functions, synced with Chargebee and HubSpot.In the customer’s cloudIngestionParsingRollupsBillingForwardersAzure, AWS logsBlocklist syncCloud firewallsSyslogTCP, UDP, TLSManagersPanorama, othersKinesisCheckpointedParsers30 vendorsOne pipelineValidate, decodeActionsAllow, deny, dropWorkersFour, on KafkaFrameworkReusable rollupsBilling in GoStep FunctionsChargebeeAnd HubSpot
Celerium: Firewall logs in, threat data the product can trust outFive parts, left to right. Cloud integrations: log forwarders and blocklist sync that a customer deploys in Azure or AWS. Ingestion: syslog listeners, a listener for firewall managers, and Kinesis consumers on parallel shards with Redis checkpoints. Parsing: 30 vendor parsers on one shared pipeline, and a catalog that normalizes firewall actions. Rollups: four Kafka workers on a reusable framework. Billing, beside them: ported to Go on Step Functions, synced with Chargebee and HubSpot.In the customer’scloudIngestionParsingRollupsBillingForwardersAzure, AWS logsBlocklist syncCloud firewallsSyslogTCP, UDP, TLSManagersPanorama, othersKinesisCheckpointedParsers30 vendorsOne pipelineValidate, decodeActionsAllow, deny, dropWorkersFour, on KafkaFrameworkReusable rollupsBilling in GoStep FunctionsChargebeeAnd HubSpot

Also at this job

Scoping
Turned a request to block domains on about 20 firewall vendors into six work items and a 16 engineer-week estimate with a stated range.
Onboarding
Sensors are provisioned automatically when an account is created.

May 2023 – Mar 2024

Principal Software Engineer, Platform

Yummly, San Carlos, CA

Smart cooking app. Lead Go developer and architect.

  • Cut operating costs 20% by moving a startup cache from Clojure to Go, running old and new side by side before cutover.
  • Built the service behind custom recipes with the mobile and web teams.
  • Designed and built CI/CD for every Go application, and a template for new Go services.

Go, AWS, PostgreSQL, Datadog, CircleCI

See the system 3 parts

Go services behind a cooking app

Pick a part to see what I did on it.

Yummly: Go services behind a cooking appThree parts, left to right. Delivery: a template for new Go services and CI/CD for every Go application. Go services: the custom recipes service and a startup cache moved from Clojure to Go. In the app: custom recipes work with the meal planner, shopping list and collections.DeliveryGo servicesIn the appTemplateNew Go servicesCI/CDEvery Go appCustom recipesPrivate, reusableStartup cacheClojure to GoMeal plannerAnd shopping listCollectionsSaved recipes
Yummly: Go services behind a cooking appThree parts, left to right. Delivery: a template for new Go services and CI/CD for every Go application. Go services: the custom recipes service and a startup cache moved from Clojure to Go. In the app: custom recipes work with the meal planner, shopping list and collections.DeliveryGo servicesIn the appTemplateNew Go servicesCI/CDEvery Go appCustom recipesPrivate, reusableStartup cacheClojure to GoMeal plannerAnd shopping listCollectionsSaved recipes

Also at this job

Sequencing
Argued for extracting the costly cache first and measuring it before splitting into microservices. We did it in that order.
First week
Simplified a metadata-service query; the planner’s cost estimate fell about 94%.
Hiring
Helped interview engineers.

Sep 2021 – May 2023

Software Engineer Manager, Connectivity

Redeam, Boulder, CO

SaaS for tours, attractions and activities.

  • Cut booking page loads from 8 seconds to 2, and raised booking performance 80%, with an event-invalidated cache.
  • Designed and built a pricing engine every system shares, delivered a week early for the Disney onboarding.
  • Onboarded 300+ offline Disney resellers through a reseller portal I designed and built.
  • Led a small team, part of it in Ukraine, while shipping or supporting seven products in six months.

Go, GCP, Helm, Vue, NestJS, PostgreSQL, Buildkite, Datadog

See the system 4 parts

Pricing, reselling and booking, made faster

Pick a part to see what I did on it.

Redeam: Pricing, reselling and booking, made fasterThe pricing engine sits in the platform as a shared library: the core API and the operator portal’s inventory API both use it, and the inventory API writes rate cards into it as price rules. The reseller portal talks to its own API, which uses the booking cache; the hosted site’s booking engine reads the same cache, and the Things To Do feed from Google lands people on that booking engine.PortalsPlatformOperatorsTours, activitiesOperator portalRate cardsReseller portal300+ resellersCore APIOperators connectInventory APIOperator portalPortal APIReseller portalPricing engineA shared libraryBooking cache8 s down to 2 sThings To DoGoogle data feedBooking engineOur hosted site
Redeam: Pricing, reselling and booking, made fasterThe pricing engine sits in the platform as a shared library: the core API and the operator portal’s inventory API both use it, and the inventory API writes rate cards into it as price rules. The reseller portal talks to its own API, which uses the booking cache; the hosted site’s booking engine reads the same cache, and the Things To Do feed from Google lands people on that booking engine.PortalsPlatformOperatorsTours, activities→ Core APIOperator portalRate cards→ Inventory APIReseller portal300+ resellers→ Portal APICore APIOperators connect→ Pricing engineInventory APIOperator portal→ Pricing enginePortal APIReseller portal→ Booking cachePricing engineA shared libraryBooking cache8 s down to 2 sThings To DoGoogle data feedBooking engineOur hosted site→ Booking cache

Also at this job

The team
Performance reviews, growth plans and mentoring.
City-pass site
Built the pass-usage API and confirmation pages of a white-label site where a city-pass partner’s customers validate a pass and get a same-day barcode for an attraction.

Jun 2020 – Sep 2021

Software Engineer, Platform

USAA, Plano, TX

Financial services for the military community.

  • Developed the mass-communication application and web app that sends messages to customers.

Java, Spring, OpenShift, React, Kafka, Cypress

May 2019 – Jun 2020

Senior Software Engineer, Connectivity

Redeam, Boulder, CO

SaaS for tours, attractions and activities.

  • Designed and built the Reserve With Google Integration, so people book connected partners from Google Search.
  • Built the operator portal for bookings, inventory, pricing and resellers.
  • Migrated a Java API to Go in increments, with a unit, integration and performance suite to prove it.

Go, Java, GCP, Helm, Vue, NestJS, Hasura, PostgreSQL, Buildkite, Datadog

See the system 2 parts

Operators’ inventory, bookable from Google Search

Pick a part to see what I did on it.

Redeam: Operators’ inventory, bookable from Google SearchOperators connect to the core API. The operator portal talks to its inventory API, which publishes inventory and availability to the core API. From the platform, the Reserve With Google Integration feeds Google Search, where people book.PlatformOperatorsTours, activitiesOperator portalManage bookingsand inventoryCore APIOperators connectInventory APIOperator portalReserve WithGoogleIntegrationGoogle SearchBook in place
Redeam: Operators’ inventory, bookable from Google SearchOperators connect to the core API. The operator portal talks to its inventory API, which publishes inventory and availability to the core API. From the platform, the Reserve With Google Integration feeds Google Search, where people book.PlatformOperatorsTours, activities→ Core APIOperator portalManage bookingsand inventory→ Inventory APICore APIOperators connect→ Reserve With GoogleInventory APIOperator portalReserve With GoogleIntegrationGoogle SearchBook in place

Also at this job

Java to Go
Each increment shipped with its documentation and sample requests. QA and I built the test suite that proved it, and monitors went in after cutover.

Earlier roles

  • Dec 2018 – Jan 2020Freelance Full-Stack Developer, Upwork. An e-commerce site on AWS and Stripe, end to end.
  • Jun 2017 – May 2019Senior Programmer/Analyst, Platform, DerbySoft, Addison, TX
  • Mar 2017 – Jun 2017Java Developer, Blue Cross Blue Shield, Richardson, TX
  • Sep 2015 – Mar 2017Java Developer, MedSynergies
  • Jul 2014 – Sep 2015Java Developer, Fidelity Investments, Southlake, TX
  • May 2013 – Aug 2013Software Engineering Intern, Cerner, Kansas City, MO

Projects

Built on my own time. Both repos are private, so they are described here, not linked.

finhub

A personal finance hub I run at home: a Go GraphQL API, a Vue app, Postgres, and a worker that syncs banks and runs the scheduled jobs. Categorization uses Claude on Amazon Bedrock, and what goes to the model is redacted first.

Ports and adapters in Go, tests at every layer, and a written design system. This page reuses that system and its diagram engine.

Go, GraphQL, Vue, PostgreSQL, Docker, Amazon Bedrock

How finhub runsThe browser talks to web (nginx and the Vue app), which passes requests to the api (Go, GraphQL). The api sends only redacted text to Claude on Bedrock. The api and the worker share Postgres, and the worker syncs banks through Lunch Flow.BrowserLaptop or phonewebnginx, Vue appClaudeBedrock, redactedapiGo, GraphQLworkerGo, scheduled jobsPostgresOne databaseBank syncLunch Flow
How finhub runsThe browser talks to web (nginx and the Vue app), which passes requests to the api (Go, GraphQL). The api sends only redacted text to Claude on Bedrock. The api and the worker share Postgres, and the worker syncs banks through Lunch Flow.BrowserLaptop or phonewebnginx, Vue app→ apiClaudeBedrock, redactedapiGo, GraphQL→ PostgresworkerGo, scheduled jobs→ Bank syncPostgresOne databaseBank syncLunch Flow

aws-platform

The shared base of my AWS account, written as Terraform: state, sign-in, the audit trail, the budget, the domain, the network, and the rules for who may deploy what. It serves this page.

GitHub Actions deploys through OIDC with no stored keys, and a project cannot change the platform it runs on.

Terraform, GitHub Actions, S3, CloudFront, Parameter Store

How this page is publishedA push to main starts GitHub Actions, which builds the site and takes a deploy role through OIDC, with no stored keys. It looks the bucket and distribution up in Parameter Store, copies the build to S3, and CloudFront serves it to you.The AWS account, as TerraformPush to mainThis repoGitHub ActionsBuilds the siteParameter StoreWhere things areDeploy roleOIDC, no keysCloudFrontServes it to youS3 bucketThe built page
How this page is publishedA push to main starts GitHub Actions, which builds the site and takes a deploy role through OIDC, with no stored keys. It looks the bucket and distribution up in Parameter Store, copies the build to S3, and CloudFront serves it to you.The AWS account, asTerraformPush to mainThis repoGitHub ActionsBuilds the site→ Deploy roleParameter StoreWhere things areDeploy roleOIDC, no keys→ S3 bucketCloudFrontServes it to youS3 bucketThe built page

Skills

Languages and frameworks
Go, Java, TypeScript, Python, Vue, NestJS, React, Spring
Platforms and data
AWS, GCP, Azure, Kubernetes, Terraform, Docker, Kafka, Kinesis, Redis, PostgreSQL, GraphQL
Practices and tools
Software architecture and design, CI/CD, Leading and mentoring engineers, Claude, GitHub Copilot, Cursor

Education

  • Aug 2018 – May 2020MS, Software Engineering, University of Texas at Dallas
  • Aug 2010 – May 2014BS, Software Engineering, University of Texas at Dallas