CV

AI engineer and applied scientist with a PhD in AI and 10 years of turning NLP and LLM research into production systems. I currently lead the multi-agent AI platform at bexio, have consulted for venture-backed AI startups, and teach LLMs and AI agents at Lucerne University of Applied Sciences (HSLU).

Experience

1. Aug 2025 – Present Rapperswil, Switzerland

   Senior AI Engineer

   bexio AG

   • Technical lead and applied scientist for bexio Intelligence. Set the technical strategy and architecture of a production multi-agent platform and took it from lab prototype to a core 2026 company OKR, with a tiered rollout to 100,000+ SME customers.
   • Architected and deployed high-concurrency multi-agent pipelines that turn complex, non-linear financial workflows into automated execution, using low-latency semantic caching to keep production compute costs under control.
   • Designed and built a plug-and-play evaluation platform that stress-tests multi-turn reasoning traces, conversation safety and workflow logic in simulation, so every agent change is validated before it reaches production.
   • Bridged engineering and cross-functional leadership. Partnered with Product, SRE and Sales to turn ambiguous customer requirements into production-grade AI guardrails and a robust system architecture.

   • Multi-agent systems
   • Technical strategy
   • Agent evaluation
   • Semantic caching
   • AI guardrails
2. Apr 2025 – Jul 2025 New York, USA (remote)

   Applied AI Engineer (Consultant)

   Valence

   • Consulted for an AI coaching startup that has since raised $50M+ in funding.
   • Engineered and shipped production LLM systems that distil long histories of user–AI conversations into high-quality personalized context for long-term goal tracking.
   • Raised product quality and user experience with personalized, real-time coaching insights, which increased user retention and improved conversation quality.

   • Production LLM systems
   • Personalization
   • Long-term memory
3. Jan 2024 – Mar 2025 Zürich, Switzerland

   Senior Research Engineer, NLP/IR

   Telepathy Labs

   • Owned the retrieval stack of a RAG-based conversational product search system. Set up a reliable retrieval evaluation, introduced document expansion for dense retrieval, and merged re-ranking with post-filtering, improving answer quality and latency.
   • Led the company’s move from pay-per-call closed-source LLMs (OpenAI) to self-hosted open-source models, cutting cost and latency while matching quality with fine-tuned LoRA adapters.
   • Founded and co-supervised a company-wide internship program (2 interns since summer 2024).
   • Drove the research behind two publications (SciChat@EACL’24 and COLING’25).

   • RAG
   • Dense retrieval
   • LoRA / PEFT
   • Open-source LLMs
   • LLM cost optimization
4. Jun 2023 – Dec 2023 Zürich, Switzerland

   Research Engineer (Intern), NLP/IR

   Telepathy Labs

   • Built an LLM-based user simulator to evaluate task-oriented dialogue systems automatically (SciChat@EACL’24).

   • Dialogue systems
   • User simulation
   • LLM evaluation
5. Mar 2020 – Jan 2024 Lugano, Switzerland

   PhD Candidate in NLP/IR

   USI Lugano, IDSIA

   • Defined and led a research agenda on mixed-initiative conversational search: generating clarifying questions and using LLM-based user simulation to evaluate conversational systems. Published at SIGIR, WSDM and CHIIR, among others.
   • Research visit to the IAI group, University of Stavanger, Norway, with Prof. Krisztian Balog (CHIIR’24).
   • Research visit to GRILL, University of Glasgow, UK, with Prof. Jeff Dalton (SIGIR and preprint).

   • Conversational search
   • Clarifying questions
   • User simulation
6. Jul 2019 – Dec 2019 Singapore

   Research Associate, NLP

   StatNLP, Singapore University of Technology and Design

   • Document-level entity relation extraction with Prof. Wei Lu (ACL’20).

   • Relation extraction
7. Oct 2018 – Jun 2019 Heidelberg, Germany

   Research Associate, NLP

   Heidelberg Institute for Theoretical Studies (HITS)

   • Deep learning methods for predicting mental health status from social media, with Prof. Michael Strube (W-NUT@EMNLP’19).

   • Deep learning
   • Text classification
8. Aug 2017 – Jul 2018 Zagreb, Croatia

   Data Analyst

   ReversingLabs

   • Data Quality / Program Management team: kept large and diverse datasets consistent and clean.

   • Data quality
9. May 2016 – Sep 2017 Zagreb, Croatia

   Student Project Associate

   TakeLab, University of Zagreb

   • Improved customer experience with NLP by predicting users’ sentiment, emotions and actions.

   • Sentiment analysis
   • NLP

Teaching & talks

1. 2 editions Rotkreuz, Switzerland

   Lecturer, CAS LLMs and AI Agents

   Lucerne University of Applied Sciences and Arts (HSLU), School of Computer Science and Information Technology

   • Lecturer in a Certificate of Advanced Studies (15 ECTS) on building production-ready LLM systems, aimed at professionals and managers.

Plus conference presentations in Norway, Spain and Germany.

Education

1. 2024 Lugano, Switzerland

   PhD in Informatics

   USI Lugano, IDSIA

   • Dissertation: Clarification and user simulation in mixed-initiative conversational search, supervised by Prof. Fabio Crestani.
2. 2018 Zagreb, Croatia

   MSc in Computer Science

   University of Zagreb, Faculty of Electrical Engineering and Computing (FER)

   • Thesis: Text-based mental disorder prediction from online discussion texts, supervised by Prof. Jan Šnajder.
3. 2015 Zagreb, Croatia

   BSc in Computer Science

   University of Zagreb, Faculty of Electrical Engineering and Computing (FER)

   • Thesis: Automatic extraction of verb relations in Croatian, supervised by Prof. Jan Šnajder.

Awards & service

• Publications at SIGIR, ACL, WSDM, ACM TIST, ECIR, COLING, CHIIR and EACL workshops (700+ citations, h-index 10). Full list on Google Scholar.
• Program committee member for SIGIR, WSDM, ECIR, CIKM and EACL

Skills