SOFTWARE DEVELOPER · AI & MULTI-AGENT SYSTEMS
I build AI agents and Python applications that execute tools, check results, and reuse experience — alongside knowledge-graph decision support and evidence-grounded troubleshooting systems.
CURRENTLY
Research Assistant Intern at the University of Helsinki — building VeriEvo for verified engineering optimization.
BASED IN
Solna, Stockholm
OPEN TO
Software development and AI engineering roles
Research Assistant Intern
Built all modules of VeriEvo, an AI agent framework that generates and improves engineering solutions through code execution, simulation, constraint checks, and reusable experience. Integrated ShipD hull optimization and reduced total resistance by 10.27% on average across 50 tasks.
Master Thesis Intern — AI & Multi-Agent Systems
Built a general-purpose multi-agent framework that turns requirements and domain knowledge into evidence-backed recommendations. Used AG2 and Neo4j to coordinate review and refinement while preserving proposals, disagreements, revisions, and final decisions for inspection.
AI Engineer Intern
Built an AI troubleshooting prototype with a FastAPI backend, Gradio interface, user sessions, and conversation history. Developed LangGraph workflows and a RAG pipeline that combined logs, technical documents, and KPIs to investigate faults with relevant evidence.
AI Evaluator & Senior Reviewer
Ranked and refined LLM outputs for code generation, reasoning, and instruction following to support RLHF fine-tuning. Audited 100+ tasks weekly with over 95% quality compliance while mentoring junior annotators.
M.Sc. ICT Innovation · Dual degree · Stockholm & Helsinki
B.Eng. Robotics & B.Sc. Computer Science · Dual degree · China
GraphRAG · Multi-Agent · Finnish Tax Law
Production-ready 6-agent GraphRAG system over a typed knowledge graph with 994K nodes. Agents decompose questions, traverse relationships, generate cited answers and self-verify.
70% key facts · 90% citations
Agent Evaluation · Iterative Refinement
A lightweight Code Agent evaluation framework measuring correctness, self-correction, token efficiency and cost with sandboxed execution and multi-model comparison.
87% pass · 60% recovery
Multi-Agent · LangGraph · RAG
Role-based multi-agent orchestration for dynamic dialogue routing, combining operational data for RCA-style troubleshooting with grounded explainability.
Multi-source · Traceable
06 / CONTACT
Open to software development and AI engineering roles focused on agents, backend systems, and applied machine learning.
Get in Touch