I work as an independent AI & ML Engineer and Consultant, taking on client engagements in various fields of machine learning, such as LLMs, Agentic AI, GenAI, NLP, Computer Vision, Traditional ML, TTS and ASR. Below is a fuller account of my work than my CV allows — including engagements I can only describe in general terms.
AI & ML Engineer and Consultant — Self-employed (Remote)
October 2023 – Present
I take on both hands-on engineering — training AI/ML models, writing production code, designing and deploying pipelines to production — and the work that comes earlier: assessing whether AI is the right tool at all, shaping architecture, and setting timelines that survive contact with reality. Some clients are under NDA and are described by engagement type rather than name.
Senior AI Engineer — Bug Triage & Root Cause Analysis Agent (Confidential Client)
2026
- Delivered a functional proof-of-concept bug triage and root cause analysis pipeline, chaining a no-code Atlassian Rovo agent into a prompt-driven GitHub Copilot agent to surface likely root causes from bug reports and shorten manual triage
- Designed an alternative system architecture in response to revised client IT policy, making development of the agent feasible
- Partnered directly with stakeholders to gather requirements and demo iterations, and received IT policy updates directly from the downstream end client
Senior AI Engineer — Enterprise Chatbot on Google Gemini (Confidential Client)
2026
- Extended a Google Gemini Enterprise chatbot with custom integrations beyond the platform’s built-in connectors, enabling the agent to apply labels to emails in Gmail and move files in Google Drive
- Evaluated the platform’s no-code and high-code capabilities and partnered with stakeholders to scope feasible features aligned with their needs
Newfire Global Partners — Senior Machine Learning Engineer
2025 – Present
- Reduced BERT-based model inference time by roughly 50% by implementing an encoder layer-pruning technique across the production and model training frameworks
- Built an NLP model end-to-end, designing the dataset via entity prelabeling and LLM-assisted labeling, training it, and integrating it into production to power a new feature that flags abnormal records
- Retrained multiple NLP models under strict validation standards, running differential testing to verify performance before release
- Developed new features and maintained production NLP pipelines within a 7+ year-old spaCy codebase, using an in-house FlairNLP-based framework for model training, and updated Jenkins CI/CD
AI Lead — Computer Vision Advertising Project (Confidential Client)
2025
- Defined the technical architecture and established realistic delivery timelines with the CEO, tech lead, and infrastructure lead
- Identified that the existing edge hardware could not run the AI pipeline, prompting a hardware change before development committed to an unworkable target
- Built the project’s first complete ML processing pipeline, running all features end-to-end where the previous prototype could not, and establishing a working baseline the incoming AI Lead reused
- Mentored and managed a university-level engineer, assigning work, providing technical direction and code review
Read the full write-up in the case study.
GoodAI, AI People — Machine Learning Engineer
2023 – 2025
- Engineered prompt-protection mechanisms that increased time-to-extract the system prompt from under 1 minute to roughly 30 minutes for technical testers, and to days for non-technical QA
- Contributed to a QLoRA training framework, adding OpenAI fine-tuning support, then inherited and used it to fine-tune local 7B LLMs (Llama, Mistral) after the lead departed
- Ran and evaluated local LLM and ASR inference using llama.cpp, whisper.cpp, and NVIDIA In-Game Inferencing (NVIGI), identifying an SDK bug that NVIDIA later patched
- Integrated local and cloud ASR/TTS, implementing a self-contained FastAPI + PyInstaller deployment for a local TTS model released under a week prior, and generated arbitrary synthetic voices by manipulating voice embeddings in vector space
- Built content-filtering and safety mechanisms plus credit-calculation logic reconciling in-game credits against actual LLM, ASR, and TTS spend, surfacing a billing discrepancy
See the live demo where I appear as “Developer Mislav,” or read the full case study.
Beyond client work: I’ve delivered lectures at AI Center Lipik, helping train the next generation of AI developers in Croatia. I’ve also spoken publicly on AI beyond engineering audiences — an invited talk on preserving cultural heritage in the age of AI, covered by Plava vinkovačka.
Machine Learning Engineer — TIS Group (Remote)
October 2020 – September 2023
My first full-time ML role, spanning healthcare and agriculture. I moved from model training and optimization on SENDD into owning a pipeline end-to-end on Vineyard Angel — design through deployment.
- SENDD (Healthcare): Reduced pose-estimation model training time from weeks to days by moving training onto GPU, and later multi-GPU, using NVIDIA’s trt_pose
- Improved pose-estimation accuracy through transfer learning and built a heatmap-feature classifier using kNN and SVM with PCA for dimensionality reduction
- Vineyard Angel: Led development of a drone-imagery pipeline from design to deployment, segmenting vineyard rows and gaps and estimating vine-gap counts with PyTorch Segmentation Models and OpenCV
- Automated the pipeline on Azure via systemd-orchestrated VMs, and presented the project with the agricultural partner at AI2FUTURE 2022
- Scoliosis estimation (hackathon): Built a system for estimating scoliosis severity — see my presentation
- CatAIog (hackathon): Built, with a team, an LLM-based chatbot over PDF documents — see my presentation
Software Engineering Intern — Rimac Automobili (Sveta Nedelja, Croatia)
July 2019 – February 2020
An extended internship on an autonomous-driving team, working part-time around my university schedule.
- Built C++ infrastructure around ML models within an autonomous-driving team, including systems parsing
low-level radar message bytes - Worked with Linux, Git, and NVIDIA autonomous-driving hardware under the mentorship of a senior engineer
Skills & Tools
Proficient in
- Domains: GenAI, LLMs, NLP, Computer Vision, Traditional ML
- Programming Languages: Python
- Libraries: PyTorch, Hugging Face, scikit-learn, FastAPI, NumPy, OpenCV
- Tools & Cloud: Azure, Git, Jupyter, Linux
Familiar with
- Domains: Agentic AI, Data Science, RAG, ASR, TTS
- Programming Languages: C++, SQL
- Libraries: FlairNLP, spaCy, pandas, LlamaIndex, Streamlit, Keras, SciPy, PyInstaller, TensorFlow
- Tools & Cloud: CI/CD pipelines, Jenkins, W&B, GCP, Vertex AI (incl. ADK), Docker, GNU Make, Terraform, Prodigy
- Databases: CosmosDB, Chroma, PostgreSQL