← Back to site
CV Version:

Kristian Garza

AI SOFTWARE ENGINEER BERLIN, GERMANY kj.garza@gmail.com

Profile

Curiosity-driven AI engineer who rapidly prototypes and ships: built open-source LLM metadata translators, scalable Embeddings & RAG pipelines, and a Rust API clustering millions of vectors in seconds. Skilled in LangChain, Hugging Face, and pgvector, I design experiments, build evaluation pipelines, and turn new research into practical PoCs—communicating results clearly to guide product strategy.

Employment History

AI SENIOR SOFTWARE ENGINEER at Digital Science, Berlin
2024— Present
  • Launched a FastAPI microservice that converts natural-language queries into optimized Dimensions searches using LLM entity extraction and pgvector, deployed on Docker/Kubernetes with automated CI.
  • Delivered a Rust-based Embeddings API that clusters research-document embeddings in seconds with parallel K-means and LLM summarization, deployed as a scalable Kubernetes microservice.
  • Built a FastAPI service using OpenAI LLMs and Redis caching to generate instant TL;DRs and key points from research articles, deployed at scale with CI/CD and monitoring.
  • Built a GitHub org catalogue scanning 1,000+ repos via a single LLM call per repo; exposed semantic search through an MCP server at ~$1 per full org scan.
  • Designed and shipped two generations of a research knowledge graph system: a Graphology-based production pipeline (7,068 entities, 19,085 relationships) and a hexagonal-architecture rewrite using Effect-TS, PostgreSQL+pgvector, and a 22-stage extraction pipeline with graph topology metrics.
  • Built an AI-powered manuscript review environment in Next.js with Tiptap v3, real-time citation verification via MCP, contradiction detection through the Dimensions API, and Overleaf sync.
  • Prototyped an internal knowledge graph product unifying Salesforce, Drive, Confluence, Zendesk, Slack, and GitHub data; led discovery, wrote the PRD, and validated architecture through three interactive demo surfaces.
PRODUCT DESIGNER at DataCite, Berlin
2020 — 2023
  • Created and open-sourced Parrot GPT—a GPT-powered Python toolkit that auto-converts and enriches publishing metadata across 20+ schemas through an extensible interface, chunk-safe processing, and CI-backed test coverage—slashing manual conversion time for libraries and research institutions.
  • Spearheaded DataCite’s first unified Design System, shipping a Bootstrap-driven, WCAG-compliant component library and Storybook docs that standardize UX across all web products and accelerate developer adoption.
FULL STACK DEVELOPER at DataCite, Berlin
2016 — 2020
  • Developed and deployed a React / Next.js frontend featuring GraphQL-powered multi-entity search, interactive relationship graphs, and high-performance SSR—complete with TypeScript, Cypress test suites, and GitHub Actions CI, elevating DataCite Commons’ research-output discovery experience.
  • Designed and delivered a Rails API that ingests, validates, and stores large SUSHI usage reports in S3/MySQL at 50K-dataset scale with JWT security and on-the-fly compression—establishing a standards-compliant, performant backbone for DataCite research-metrics tracking.
PHP DEVELOPER at DataMine
2011 — 2012
  • Devised features for an ERP system using a Symfony-like framework in PHP, increasing system efficiency and significantly enhancing maintainability due to the implementation of modern programming practices. Tools: PHP, MySQL.
Visit linkedin.com/in/kjgarza for full EMPLOYMENT HISTORY

Education

PHD COMPUTER SCIENCE, University of Manchester, Manchester
2012 — 2016

Led the investigation of the employment of a novel choice architecture approach to integrate the captured context into the data repository design, developing features that resonated with the user base.

MSC SPACECRAFT TECHNOLOGY, University College London, London
2007 — 2008

Defined design improvements of electron detectors.

Recent Hacks

Converting Publications into Interactive Podcasts, Holtzbrinck Hackathon 2024
2024

Co-led a 48-hour sprint to ship a Next.js app that transforms academic PDFs into multi-speaker, podcast-style episodes with live Q&A, marrying custom LLM condensation and speech-synthesis pipelines (TTS) with an accessible, mobile-first UI.

Workflow Automation for Research Communities, Holtzbrinck Hackathon 2025
2025

Build a no-code workflow-automation platform in a Next.js app that connects research staples such as Dimensions, Overleaf, and ReadCube, letting scholars stitch together cross-tool integrations in minutes and eliminating brittle, ad-hoc scripts. services through a no-code interface.

Visit website for OTHER PROJECTs

Published Work

Revolutionizing Metadata Schema Mapping with ChatGPT and AI. , Substack
2023 — https://doi.org/10.59350/b9na4-hq881
ParrotGPT: On the Advantages of Large Language Models Tools (AI) for Academic Metadata Schema Mapping. , EOSC
2023 — https://doi.org/10.59350/hs9k1-wn031

Courses

Designing AI Experiences, NN/Group
2025 — https://www.nngroup.com/courses/designing-ai-experiences/
Designing Complex Apps for Specialised Domains, NN/Group
2022 — https://www.nngroup.com/courses/complex-apps-specialized-domains/

Kristian Garza

TECH LEAD · AI ENGINEER BERLIN, GERMANY kj.garza@gmail.com

Profile

Senior AI Engineer and technical leader with 10+ years building production software and 4+ years shipping LLM-powered systems at scale. At Digital Science I architected Python/FastAPI RAG microservices—agentic orchestration, vector-search APIs, LLM evaluation—deployed on Kubernetes/AWS. At DataCite I led a 4-person cross-functional team to deliver a WCAG-compliant design system and shaped product strategy through rigorous user research. PhD in Computer Science, 18+ publications on LLMs and scholarly metadata, and deep experience in responsible AI for knowledge-intensive domains.

Employment History

AI SENIOR SOFTWARE ENGINEER at Digital Science, Berlin
2024 — Present
  • Architected and shipped a Python/FastAPI RAG microservice that translates natural-language researcher queries into structured Dimensions searches via LangChain entity extraction and pgvector semantic search—deployed on Kubernetes with Sentry/Prometheus observability.
  • Built a Python/FastAPI service using OpenAI LLMs and Redis caching to generate real-time article summaries and key-point extractions from research literature at production scale with full CI/CD.
  • Delivered a Rust-based embeddings API clustering research-document vectors with parallel K-means and LLM summarization—designed as a scalable Kubernetes microservice serving live traffic.
  • Facilitates AI Technology Radar sessions to evaluate and adopt emerging GenAI tools (LLM evaluation frameworks, agentic orchestration, vector stores) across the engineering organisation.
PRODUCT DESIGNER & ENGINEERING LEAD at DataCite, Berlin
2020 — 2023
  • Led a 4-person cross-functional team (designer, developer, PM, design manager) to design and ship DataCite's unified Design System—Atomic Design component library, WCAG-compliant Storybook docs, Bootstrap-compatible JS package—used across all web products.
  • Owned product strategy for a harvesting service relaunch: ran a 56-person focus group across 10 organisations and an 85-response survey, facilitated design sprints that cut delivery lead time by 42% and opened 4 new customer relationships.
  • Created and open-sourced Parrot GPT—a Python toolkit using GPT-3/3.5 for automatic bibliographic metadata translation across 20+ schemas—an early production LLM tool for knowledge management pipelines.
FULL STACK DEVELOPER at DataCite, Berlin
2016 — 2020
  • Designed and delivered Sashimi, a Rails API implementing the SUSHI protocol for research usage metrics at 50,000-dataset scale—S3/MySQL hybrid storage, JWT auth, and on-the-fly compression reduced storage costs by 70%.
  • Built DataCite Commons: a React/Next.js frontend with GraphQL-powered multi-entity search, interactive relationship graphs, and SSR—with TypeScript and Cypress test suites.
Visit linkedin.com/in/kjgarza for full EMPLOYMENT HISTORY

Education

PHD COMPUTER SCIENCE, University of Manchester, Manchester
2012 — 2016

Investigated novel choice-architecture approaches for data-repository design, delivering user-centric features grounded in controlled experiments and contextual inquiry.

MSC SPACECRAFT TECHNOLOGY, University College London, London
2007 — 2008

Defined design improvements for electron detectors.

Recent Hacks

Workflow Automation for Research Communities, Holtzbrinck Hackathon 2025
2025

Built a no-code workflow-automation platform connecting research tools (Dimensions, Overleaf, ReadCube) via a Next.js app—letting scholars stitch cross-tool integrations in minutes, eliminating brittle ad-hoc scripts.

Converting Publications into Interactive Podcasts, Holtzbrinck Hackathon 2024
2024

Co-led a 48-hour sprint to ship a Next.js app transforming academic PDFs into multi-speaker podcast episodes with live Q&A, using custom LLM condensation and TTS synthesis pipelines.

Visit website for OTHER PROJECTs

Published Work

The Impact of Language User Interfaces on Finding Scholarly Repositories. , iPRES 2023
2023 — https://doi.org/10.59350/b9na4-hq881
ParrotGPT: On the Advantages of Large Language Models for Academic Metadata Schema Mapping. , EOSC 2023
2023 — https://doi.org/10.59350/hs9k1-wn031
Academic Publishing Web Forms Meet Your Demise: The Unstoppable Rise of Large Language Models. , Substack / Force11
2023 — https://doi.org/10.59350/b9na4-hq881

Courses

Designing AI Experiences, NN/Group
2025 — https://www.nngroup.com/courses/designing-ai-experiences/
Designing Complex Apps for Specialised Domains, NN/Group
2022 — https://www.nngroup.com/courses/complex-apps-specialized-domains/