Home / About
I design multi-agent workflows that follow rules, pause for a person when they should, and keep their evidence beside every claim.
I'm an Information Systems undergraduate at Singapore Management University, expected to graduate in May 2027. In 2026 I began work as an AI Solutions Intern.
My recent work is on AI systems that have to stay inside rules. I owned the Clarifier and Orchestrator coordination layer of a team's multi-agent wealth-advisory prototype. I co-developed a corporate-onboarding proof of concept and contributed to its review-and-recovery workflows. In my internship I contributed the integration work for a local computer-vision review prototype.
Internship
Contributing to SafeTrace, a team prototype for reviewing visual safety evidence locally: the React/FastAPI integration, local workflow hardening, the evidence-review experience and a hosted-to-local connector.
SafeTrace case studySingapore Management University · expected graduation May 2027
Singapore Polytechnic
Grouped by what I've done, each with the project or certificate that shows it. Tools are listed only where I used them in my own part of the work.
Rule-first intent parsing, deterministic plan admission and single-use approval steps around a team's specialist agents.
Python · FastAPI · LangGraph
finAIBuilt by me
An evaluation harness for a coordination layer, and automated tests for a browser-to-local boundary and for retrieval with provenance.
Workflows where an exception pauses a case for a person, and the case resumes from a saved checkpoint once it's resolved.
Corporate OnboardingContributed · team POC
React interfaces on FastAPI backends, local job and batch processing, and a connector from a hosted page to a local backend.
React · FastAPI · Python
SafeTraceContributed
Retrieval, scope resolution and provenance, so answers stay tied to the material they came from.
LoreDropContributed
Python programming, data visualisation and applied machine learning, through structured coursework.
Python
AI100 · AI200Certificates
Rules stay the authority; models assist inside boundaries. In finAI, deterministic rules govern authorisation, plan admission and execution, and optional model assistance can't override them.
Seen in finAI →Exceptions pause for a person instead of producing a substitute answer. In Corporate Onboarding, a conflict stops the case at the step that found it, and a reviewer resolves it before the case resumes.
Seen in Corporate Onboarding →Findings link back to their source, and limits sit beside results. In SafeTrace, every finding ties back to the frames that produced it, and the limits are stated next to the work.
Seen in SafeTrace →The best way to reach me is LinkedIn.
Selected work: finAICorporate OnboardingSafeTraceAll work