I turn ambiguous AI ideas into working agent systems with tools, retrieval, evidence, and evaluation.
My current focus is ChainRisk Agent, an evaluation-backed Web3 risk triage prototype
that separates deterministic risk scoring from LLM-generated summaries.
I am completing an engineering master's degree at the University of Sheffield after a
dual-degree background in automatic control systems. My practical work sits at the
intersection of LLM applications, RAG, backend APIs, and system safety.
For AI Agent roles, I position myself as a hands-on builder: I can break a fuzzy task
into input routing, tool calls, retrieval, deterministic checks, structured output,
logs, and evaluation cases.
Experience
RAG and applied AI work
2025.02 - 2025.08
RAG Engineer Intern, Robin AI
Worked on retrieval-augmented question answering for contract and legal-document
workflows, focusing on document preprocessing, retrieval quality, citation-aware
prompting, and failure analysis for internal evaluation.
Handled document structure concerns such as OCR text, natural sections, tables, and bullet-style content.
Explored query expansion, hybrid retrieval, reranking, and citation-aware answer generation.
Kept claims evidence-bound: public resume metrics should be backed by internal reports before being presented as production impact.
Featured project
ChainRisk Agent
Evidence-grounded Web3 risk triage agent. The system accepts a wallet, token, or
project input and returns risk_level, risk_score, evidence, tool_trace,
uncertainties, next_checks, and safety notes.
