Portrait of Bruk Getnet

Bruk GetnetPortfolio

Bruk Getnet — Full-Stack Developer, Software Engineer, AI Data Specialist

Build Scale Optimize Evaluate

Bruk
Getnet

  • Full-Stack Developer
  • Software Engineer
  • AI Data Specialist

Scroll

  1. I build systems.

    50,000+ active users on a real-time collaborative editor, grown from 21,000

  2. I scale them.

    500K+ API requests a day across containerized Node.js services

  3. I optimize them.

    <100ms response latency held under load

  4. Now I evaluate them.

    Thousands of model responses reviewed and ranked against structured rubrics

Seven years, four chapters

From writing the software to judging the software that writes.

  1. I CamundaGmbH · Cawemo Mid-Level → Senior Frontend Engineer 20192022
  2. II Expedia Group Full-Stack Developer 20232025
  3. III Revelo LLM Training & Evaluation 2025Now
  4. IV AfterQuery AI Data Specialist 20262026

Chapter I

Camunda

Cawemo — BPMN diagram editor, SaaS2019 — 2022

Mid-Level → Senior Frontend EngineerBerlin, remote

A drawing surface where dozens of people edit the same diagram at the same moment, and none of their edits are allowed to disappear.

Started at 21,000 active users

50,000+

active users across enterprise customers

Real-timecollaboration

Collaborative editing built in React.js with MobX state management, resolving simultaneous edits through WebSocket-based conflict resolution algorithms so multiple people could work on one diagram without data inconsistencies.

The bpmn-js editor was integrated with custom extensions for BPMN 2.0 model rendering, validation and serialization, tuned to stay responsive on diagrams of 1,000+ nodes.

  • React.js
  • MobX
  • WebSockets
  • bpmn-js
  • BPMN 2.0
AK MR JD
Three editors · one document Conceptual illustration of the collaborative editing problem. Not a product screenshot.

50% of the codebase, migrated

Half the codebase, moved to TypeScript

A comprehensive refactor of a mature product: 50% of the codebase migrated to TypeScript, with Storybook component documentation and DevTools set up alongside it to shorten onboarding for new engineers.

35%fewer bug incidents

40%developer productivity gain

The record at Camunda

6
major releases shipped, v1.4 — v1.9.0
0
critical production incidents
95%+
test coverage, Jest and React Testing Library
2
junior engineers mentored — both promoted to mid-level

Every kilobyte
had to justify itself.

JavaScript bundle

1.9MB

2.8MB → 1.9MB

Initial load

1.2s

4.5s → 1.2s

Achieved through code splitting, lazy loading and MobX optimization. Intermediate values are illustrative of the direction of travel; the start and end figures are the measured ones.

Chapter II

Expedia Group

Travel platform, global scale2023 — 2025

Full-Stack DeveloperDistributed systems

The unit of work stopped being a component and became a request — one of half a million a day, each expected back in under a tenth of a second.

Conceptual representation of the technologies described below — not any employer's internal architecture.

RequestClient, global
EdgeLoad balancing
APIREST · GraphQL
AuthOAuth2 · JWT · RBAC
ServiceNode.js microservice
ContainerDocker image
OrchestrationKubernetes
DataPostgreSQL · Redis
Response<100ms

500K+

API requests handled per day

<100ms

latency, sustained

60%

reduction in deployment time

Pipelines

Automated CI/CD with Docker and GitHub Actions, cutting deployment time by 60% and making releases routine rather than eventful.

  • Docker
  • GitHub Actions
  • Kubernetes
  • Load balancing

Access

Enterprise authentication with OAuth2 and JWT plus role-based access control, securing multi-tenant access to user accounts.

  • OAuth2
  • JWT
  • RBAC
  • Multi-tenant

Product

Worked with product, UX and stakeholders to take data-driven customer-facing features from idea to global launch, adopted by millions of users.

  • REST
  • GraphQL
  • Internal tooling
  • Automation

I spent seven years building systems.

Now I study how systems behave.

Chapter III

Revelo

LLM training & evaluation2025 — Present

Expert feedback for frontier modelsRLHF datasets

Models are trained on judgement. Someone has to supply the judgement — at volume, with consistency, and with technical reasoning attached.

One prompt Two candidate answers, read line by line, then ranked against the same criteria.

Response A

—rubric score

Response B

—rubric score

Preference recorded

Structured scoring rubric

  • Technical correctness
  • Reasoning quality
  • Completeness
  • Instruction adherence

Illustrative rubric structure — criteria vary by project and domain.

Volume with consistency

Evaluated and ranked thousands of AI-generated responses across diverse technical domains using structured scoring rubrics, maintaining consistency under high-volume demands.

The pipeline behind it

Built and optimized manual and semi-automated sourcing and review pipelines — identifying, onboarding and managing contributions from technical experts, with detailed technical reasoning and corrective feedback feeding directly into RLHF datasets.

Reading the failures

Analyzed response quality patterns, identified gaps in model performance and supplied high-signal data that improved training methodologies, working alongside AI research teams.

Chapter IV

AfterQuery

Advanced LLM development & evaluationJan 2026 — Jul 2026

AI Data SpecialistSonnet · Opus

Build. Scale. Optimize. Evaluate. This is where the four verbs meet: an engineer's reading of what a model got wrong, written down so it can be fixed.

  1. 01

    Prompt

    Designed and validated high-quality prompts across software engineering and technical domains.

  2. 02

    Reference answer

    Authored the answer the model should have been able to produce, so scoring had a ground truth.

  3. 03

    Rubric

    Built evaluation rubrics covering factual accuracy, reasoning quality, code correctness and instruction adherence.

  4. 04

    Review

    Reviewed model responses against those criteria across multiple projects, including Project Pluto and Project Silver.

  5. 05

    Failure pattern

    Identified weaknesses, documented recurring failure patterns and returned actionable feedback to improve model performance.

Project 01

AI Code Debugging
Assistant

  • Python
  • OpenAI API
  • React
  • TypeScript
  • FastAPI

  1. Stack trace in

    TypeError: cannot unpack non-sequence
  at services/ingest.py:142
  at pipeline/run.py:37

  2. Parse

    FastAPI endpoints read the trace, isolate the frame that matters and normalize it.

  3. Classify

    An automated error classification engine powered by NLP assigns the failure a category.

    Type error · unpacking

  4. Suggest

    The OpenAI API is used to propose a fix, surfaced in React components written in TypeScript.

50%

less average debugging time in test environments

85% code coverage — pytest and Jest, kept green in CI

Project 02

Distributed Web
Vulnerability Scanner

  • Python
  • Node.js
  • Docker
  • MySQL
  • CI/CD

1,000+

endpoints examined per minute

Detection classes

  • SQL injection
  • Cross-site scripting
  • Cross-site request forgery

A Dockerized scanning platform in Python and Node.js, built into CI/CD pipelines, auto-generating security reports. Results stored in MySQL with SQL scripts for aggregation and retrieval of security metrics. Endpoint identifiers shown here are illustrative.

Project 03

Real-Time Trading
Dashboard

  • React
  • Node.js
  • WebSocket
  • Redis
Illustrative data — this page is not connected to any market feed or trading system.

<100ms

latency for real-time market updates

  • FeedWebSocket
  • CacheRedis
  • ClientReact
  • ServiceNode.js

Real-time crypto trading analytics with live WebSocket market feeds and Redis caching, holding sub-100ms latency through an optimized data streaming architecture.

Public work

github.com/brukcodes

Loading public repository data from GitHub…

Public repository activity

Derived from the most recent push date of each public repository.

All repositories on GitHub

Working set

The toolkit

  • JavaScript (ES6+)
  • TypeScript
  • Python
  • Java
  • SQL
  • Go
Bruk Getnet, photographed in a dark suit against a grey backdrop

BrukGetnet

Full-stack developer with seven years of experience building web applications in JavaScript, TypeScript, React, Node.js and relational databases — implementing front-end features, designing back-end endpoints, and debugging across the stack, including performance tuning and CI/CD automation.

Engineering
Real-time collaborative interfaces, containerized Node.js microservices, high-throughput APIs, performance tuning and deployment automation.
AI evaluation
Prompts, reference answers and rubrics for frontier models; response review for factual accuracy, reasoning quality, code correctness and instruction adherence; preference ranking and failure-pattern documentation.
Security
Secure coding practices, OAuth2 and JWT with role-based access control, and a distributed scanner built to find SQL injection, XSS and CSRF across web applications.
Education
Bachelor of Science, Computer Science.

Adapts quickly to new technologies and works well across product, design and research teams.

Let's build what's next.

Bruk Getnet

  • Full-Stack Developer
  • Software Engineer
  • AI Data Specialist