Patrik P. Süli
Software Architect · AI Engineer · PhD Candidate · Founder of KenguTech
- Emailpatrik.suli@mailbox.org
- Universitysuli.patrik@uni-obuda.hu
- WebPatrikPSuli.com
- LinkedInpatrik-süli
- ORCID0009-0001-9481-3664
- ResearchGatePatrik-Sueli
- Google Scholarh4ASwVUAAAAJ
- LocationMartonvásár, Hungary
Profile
Software architect with a decade of professional JVM engineering, currently leading enterprise microservice systems sized for millions of requests at KenguTech, the studio I founded in 2022. PhD candidate at Obuda University and Assistant Lecturer at the John von Neumann Faculty of Informatics, with peer-reviewed research on applied large language models, agentic systems for scientific workflow, and LLM evaluation methodology.
Experience
HUN-REN Centre for Agricultural Research (ATK)
Software Architect · AI Engineer
Sep 2026 – Present
Architecting research informatics for plant breeding and crop science at the HUN-REN Agricultural Research Centre in Martonvásár: modernising the scientific software and data infrastructure and embedding LLM and machine learning capabilities into the research workflow.
KenguTech
Software Architect · Founder
Dec 2022 – Present
Founded the studio and lead time-and-materials engagements end-to-end: architecture, hands-on delivery, and operational ownership. Specialised in enterprise-grade Java microservice systems sized for millions of requests, alongside applied LLM solutions.
- Customer Portal Re-Platforming2025 – Present
Architect and lead developer on the modernisation of a public portal serving millions of utility customers. Decomposing a legacy monolith into a high-availability microservice topology aligned with modern DevOps and cloud-native practice, with reliability and operational maintainability as primary constraints.
Stack: Java · Spring Boot · Maven · REST · Microservices · PostgreSQL · Redis · RabbitMQ · Debezium · Keycloak
- High-Throughput Billing Platform2023 – 2025
Co-architect on a 25+ microservice distributed billing platform sized for millions of customers. Led the planning and specification of the high-throughput data and integration pathways underpinning the platform's correctness and performance guarantees.
Stack: Java · Spring Boot · Maven · REST · Microservices · PostgreSQL · Redis · Elasticsearch · Kafka · Debezium · Keycloak · GitLab CI · Kubernetes
- AW Server DevSecOps Pipeline2022 – 2023
Owned the GitLab CI/CD pipeline for AW Server, GE HealthCare's enterprise medical-imaging visualization platform deployed across hospital networks. Built and maintained the surrounding Bash and Python tooling, plus the vulnerability-scanning workflow that gated each release.
Stack: Python · Bash · GitLab CI · Nessus · SCAP · RHEL · SLES
Obuda University · John von Neumann Faculty of Informatics
Assistant Lecturer & Researcher
Aug 2024 – Present
Teach and supervise at the Biomatics and Applied AI Institute alongside an active research agenda. Focus areas: applied large language models, agentic systems for scientific workflow, and LLM evaluation methodology.
Freelance
Software Developer · Consultant
Jan 2018 – Nov 2022
Nearly five years of independent and contract delivery across Java (Spring Boot, Quarkus), TypeScript (NestJS, Angular), with occasional C# work. Across the engagements I shipped microservice backends, full-stack products, and the CI/CD pipelines and environments around them, often serving as the de-facto DevOps engineer too.
- Hungarian e-Health Integration Gateway2022 – 2022
Software architect and lead developer on the gateway that translated private-hospital messaging into EESZT, the Hungarian national e-Health infrastructure every healthcare provider must report to. Also led the team's day-to-day task management and delivery rhythm.
Stack: Java · Spring · Spring Boot · Maven · REST · MongoDB · Redis · Docker · GitLab CI · Proxmox
- Enterprise Conversational AI Platform2021 – 2022
Software architect and lead developer on Talk-A-Bot's enterprise conversational-AI platform, which serves brands operating under regulated-communications constraints. Built the natural-language-processing pipelines around external NLP services and the API surface that exposed them.
Stack: Java · Spring · Spring Boot · Maven · REST · Vue.js · JPA · Hibernate · PostgreSQL · Flyway · Tomcat · Docker
- Real-Time E-Learning Platform2020 – 2020
Built the Spring Boot microservices behind the WebSocket-based real-time chat layer of a synchronous e-learning platform sized for classroom-scale concurrency. Stood up the Docker-driven CI/CD pipeline that delivered them to production.
Stack: Java · JHipster · Spring Boot · Maven · JPA · Hibernate · Liquibase · Oracle · PostgreSQL · Elasticsearch · WebSocket · Docker · Kubernetes
- MÁV Rail Document Interoperability Service2020 – 2020
Spring Boot microservices that integrated MÁV's document workflows with EU rail-interoperability services, with .NET Core components handling the cross-runtime interop boundary.
Stack: Java · Spring Boot · Maven · JPA · Hibernate · Liquibase · Oracle · REST · Swagger · C# · .NET Core
- Unified Multi-Tenant HR Platform2020 – 2020
Full-stack delivery of a multi-tenant HR product: NestJS backend, Angular frontend, plus the GitLab-driven dev and test environments and pipelines around them.
Stack: NestJS · Node.js · TypeScript · Angular · MongoDB · TypeORM · Mongoose · Docker · GitLab CI
- Unified Transit API Gateway2019 – 2019
Quarkus-based aggregation service unifying multiple public-transport operators behind a single REST API, normalising the disparate provider responses into one consistent schema.
Stack: Java · Quarkus · Maven · JPA · Hibernate · Liquibase · Oracle · PostgreSQL · REST · Swagger
- ELBA Architectural CAD Platform2018 – 2019
Spring Boot microservices powering an architectural-CAD SaaS, with integrated OTP Simple payment for subscription billing. Also delivered the dev, test, and prod environments and the CI/CD pipelines that connected them.
Stack: Java · Spring Boot · Maven · Gradle · JPA · Hibernate · Liquibase · Oracle · PostgreSQL · REST · Angular · Docker
NETI IT Consulting Ltd.
Jul 2016 – Jun 2019
- Software DeveloperMar 2017 – Jun 2019
Delivered cross-platform C++ desktop applications on the Qt framework, then moved into full-stack development on the company's flagship product line with JavaFX and Spring Boot.
- Trainee DeveloperJul 2016 – Mar 2017
First professional role. Contributed production features to a graph-designer application in C++ with the Qt framework as part of the platform team.
Education
- Ph.D. · Applied Informatics2024 – 2028
Obuda University · Doctoral School of Applied Informatics & Applied Mathematics
- M.Sc. · Computer Science2022 – 2024
Eötvös Loránd University
- B.Sc. · Computer Science Engineering2013 – 2017
John von Neumann University
Technical Skills
- Languages
- Java · TypeScript · Python · C++ · C# · Bash
- Frameworks
- Spring Boot · Quarkus · Helidon · NestJS · FastAPI · JHipster · Angular · Vue.js
- Data & messaging
- PostgreSQL · MongoDB · Redis · Oracle · Elasticsearch · pgvector · Qdrant · Kafka · RabbitMQ · Debezium
- Infra & DevOps
- Docker · Kubernetes · GitLab CI · GitHub Actions · ArgoCD · Keycloak · Nessus · SCAP · RHEL
- AI / LLM
- LangChain · LangGraph · OpenAI · Anthropic · Llama · Qwen · DeepSeek · Ollama
- Testing & quality
- JUnit · Testcontainers · Pact · k6 · Gatling · SonarQube · SpotBugs
Open Source
- benchgap — Co-authorhollorol/benchgap
Fills the empty cells of LLM leaderboards. Calibrates benchmarks against each other on the models measured on both, predicts each missing score from that model's own results, and attaches a cross-validated error and confidence level to every estimate. Data refreshed daily from BenchLM, with a public OpenAPI 3.1 JSON API. Built with Roland Hollós.
- Distribird — AuthorHUN-REN-AI1Science/Distribird
Multi-agent LangGraph pipeline that turns scientific literature into informative Bayesian priors. Searches Semantic Scholar and OpenAlex in parallel, extracts reported numerical values, and fits a probability distribution via AIC model selection. Built with HUN-REN AI1Science.
- SpotBugs — Contributorspotbugs/spotbugs
6 merged pull requests to the widely-used Java static-analysis tool, including two new CERT Java detectors with full test cases: SER11-J (prevent overwriting of externalizable objects) and VNA04-J (ensure calls to chained methods are atomic). The atomicity detector connects to the published Acta Cybernetica paper on multithreading atomicity.
Selected Publications
Emergent Health Preferences in Large Language Models: A Time-Trade-Off Analysis Using EQ-5D-3L.
Automation 5.0: The Step to Systems Intelligence for a Sustainable Future.
Piloting ModiBot: A Large Language Model-Based Moderator in Normal and Emotionally Challenging Focus Group Interactions.
A Survey of Large Language Models as Research Assistants.
DeSci for Control Education: From CDIO to iCDIOS.
Augmenting virtual brainstorming workshops via agentic AI.
CoordField: Coordination Field for Agentic UAV Task Allocation in Low-altitude Urban Scenarios.
Multithreading Atomicity: Static Analysis Checkers.
Web application development for diabetes patients.
Working Papers & In Progress
benchgap: Mind the Gap: Explicit Cross-Benchmark Calibrations for Estimating Missing LLM Evaluation Scores.
Most LLMs are evaluated on only a few benchmarks: across 401 models and 146 benchmark versions, coverage is just 8.3%. benchgap estimates the missing scores without running any evaluations, calibrating each benchmark against same-capability benchmarks with monotone curves selected by leave-one-out cross-validation, and attaching an error and confidence level to every estimate. In a 15% holdout it fills 71% of masked cells at 4.6 pp MAE, 23% lower than a matrix-completion baseline. It also quantifies the harness tax, the disagreement between evaluation setups on the same models. Estimates are updated daily at benchgap.net.
SchemaEvolve: Data Model Refinement for LLM-Based SQL Code Generation.
Keeps the Text-to-SQL model fixed and optimises the database schema it has to read instead. An evolutionary loop proposes restructured schemas as code, each candidate must preserve the original data, and fitness is the execution accuracy of a small local translator. On 13 real databases from Spider 2.0-Lite, 20 generations raise the mean accuracy of three small local models from 27.1% to 50.6% at about $3.50 per database, and the two stronger evolved translators overtake Claude Opus 4.6 reading the original schemas. A local 27B mutator recovers most of the gains at zero API cost.
Distribird: Literature-Informed Prior Distribution Design for Bayesian Model Calibration.
A multi-agent LangGraph pipeline that turns scientific literature into informative Bayesian priors. Searches Semantic Scholar and OpenAlex in parallel, extracts numerical values, and fits a probability distribution via AIC model selection, improving MCMC sampling efficiency in 80% of cases across 20 parameters in ten scientific domains. Preprints are live on arXiv and Research Square; the manuscript is under review at Scientific Reports.
When Graders Drift and Models Hallucinate: Auditing Human Grading with Local LLMs.
A GDPR-compliant grading framework that runs four open-weight LLMs against teacher grades on six C# programming exams (159 submissions, 73 students). The best model reaches 75.3% subtask-level exact-match and ICC 0.974 with the human grader; the ensemble flags 122 likely teacher errors, 85% of which are stricter (not more lenient) than the model consensus. Presented at the 35th Annual Conference of the European Association for Education in Electrical and Information Engineering; proceedings publication forthcoming.
Syxplain: A Neurosymbolic Framework for LLM-Guided Symbolic Regression and Automated Scientific Discovery.
A neurosymbolic framework that pairs LLM-guided symbolic regression with automated scientific discovery, closing the loop from data to candidate equation to written explanation. Rather than only steering the search, it interprets the entire Pareto front and grounds every inference in real literature via deep formula research. P. P. Süli and R. Hollós contributed equally and share first authorship.
University Students in the Knowledge Transfer Value Chain: LLM-Mediated Knowledge Sourcing in Higher Education.
Examines how LLMs may shape knowledge sourcing, the first stage of the knowledge transfer value chain in higher education. Combines a structured literature review, persona simulation (25 student personas prompted across GPT-4o, DeepSeek, Llama 3.3 and Qwen3), and a small qualitative post-test with real students at Obuda University. The simulated students use LLMs mainly for brainstorming, information acquisition and knowledge structuring, while still preferring face-to-face interaction for collaborative learning and unfamiliar material.