Aung Zar Ni Min

Data Science Intern

RSM Singapore

Hardened and productionised a LangChain/LangGraph RAG chatbot — prompt governance, failure-mode triage, and citation integrity.

• Migrated a LangChain/LangGraph RAG chatbot's hardcoded prompts into an MLflow Prompt Registry, making prompt changes revertible and auditable without a FastAPI service redeploy.
• Conducted comprehensive post-UAT analysis, analysing feedback from over 200 test cases across 30 users before drafting targeted improvements.
• Ran a systematic UAT triage across 15 MLflow prompt-registry versions, diagnosing and fixing failure modes in the LangGraph RAG response generation (backed by Qdrant vector retrieval and Azure OpenAI via langchain-azure-ai).
• Fixed a citation-integrity bug by grounding citations with source-document filenames threaded through the retrieval nodes and hardening anti-truncation rules.
• Improved response-formatting reliability in the LangChain-driven prompt logic through extensive prompt engineering.

Applied Intelligence Intern

Income Insurance Limited

End-to-end analytical review of a production Fraud, Waste & Abuse engine — surfacing $14.4M in missed savings and shipping detection enhancements.

• Leveraged Python, SQL and Tableau to conduct an end-to-end analytical review of a production Fraud, Waste & Abuse (FWA) engine, evaluating 18 leakage hypotheses across risk categories for IncomeShield claims.
• Discovered and communicated $14.4 Million in missed savings from approved-but-not-attempted fraud flags using Tableau FIXED-LOD aggregation, correcting initial estimates caused by row-level duplication.
• Diagnosed a quarter-specific false-positive spike of 52.3% by pinpointing a single root-cause hypothesis with weak standalone precision, then profiling the false-positive population across 6 dimensions to reveal a concentrated, addressable pattern.
• Designed and implemented a Python fraud-detection enhancement on AWS, applying updated billing rules, regex-based parsing, and dataset joining to reduce false-positive flag rates, validated against internal testing data.
• Investigated and resolved model bugs affecting claim price floors, building a 22-field Tableau calculated-field pipeline to parse claims data embedded in unstructured free-text fields and quantify total dollar impact.
• Managed and pushed model enhancements and bug fixes via AWS CodeCommit, delivering production-ready changes across 4 key hypotheses.
• Drafted a tiered recommendations framework for claims managers, translating 18 leakage hypotheses' worth of dashboard analysis into prioritised, dollar-quantified action items.

IT Process Analyst Intern

Shiseido Asia Pacific

Analysed regional IT process data against SAP systems and cut Asia-Pacific infrastructure costs by up to 40%.

• Queried and analysed regional IT process data using SQL against SAP systems to identify workflow inefficiencies, producing Power BI dashboards that informed infrastructure-optimisation decisions across the company.
• Monitored Azure virtual machine and server infrastructure performance metrics, building reports that identified redundant expenditure and drove infrastructure-cost reduction across the Asia Pacific region by as much as 40%.
• Leveraged various regression models via Python to perform predictive analysis on key network and infrastructure data, identifying bottlenecks in crucial IT processes and improving efficiency.