I'm an AI/software engineer interested in building intelligent systems that work beyond the prototype.
My work spans AI agents, LLM systems, retrieval, applied ML, reinforcement learning, and full-stack AI applications — from experimenting with models and architectures to building APIs, evaluation pipelines, and production integrations.
I enjoy problems where the right solution isn't obvious at the start: understanding the problem structure, choosing the right technical approach, building the system end-to-end, and testing whether it actually works.
I've worked across 20+ projects in manufacturing, logistics, pharma, and distribution, and have built AI systems in both client-facing and enterprise environments across Korea, Malaysia, Singapore, Thailand, Brazil, and the US.
Currently exploring AI agents, AI for software engineering, retrieval & evidence-grounded reasoning, evaluation, and reliable AI systems.
An AI agent that investigates production incidents across a synthetic microservice environment.
Uses native tool calling to inspect logs, metrics, distributed traces, deployments, configuration changes, dependencies, and runbooks.
Built fault-injection scenarios for deployment regressions, dependency failures, bad configurations, connection-pool exhaustion, and traffic spikes — with evidence validation, bounded investigation loops, human approval for high-risk remediation, and stateful recovery verification.
Python FastAPI LLM Tool Calling Distributed Tracing Agent Evaluation
A research-inspired fault-localization system based on the AutoFL approach.
Given a stack trace, ranks the functions most likely to contain the bug using AST-level code chunking, hybrid keyword + embedding retrieval, and optional Claude reranking.
Achieved 100% Top-1 / Top-5 retrieval accuracy on a 5-case controlled benchmark, with automated evaluation through GitHub Actions.
Python AST Embeddings Hybrid Retrieval Claude API Evaluation
A multilingual classification system for 100K+ records, combining learned embeddings with rule-based logic and human feedback.
Built with Triplet Loss + Hard Negative Mining and a human-in-the-loop retraining workflow, achieving 95%+ classification accuracy.
PyTorch Embeddings Triplet Loss Hard Negative Mining Human-in-the-loop
A real-time voice AI system for detecting and responding to suspicious calls.
Whisper handles speech-to-text, a lightweight rule-based layer provides low-latency and explainable risk detection, and GPT is used selectively for language generation.
Built end-to-end from backend and telephony integration to frontend and deployment.
Whisper OpenAI API Flask Twilio React GitHub Actions Render
A multilingual entity-resolution system for ~50K global records with inconsistent identifiers and noisy text.
Reduced a naïve ~151M pairwise comparison space to tens of thousands of candidates through:
Blocking → ANN Retrieval → Composite Similarity Scoring
Python ANN Embeddings Entity Resolution Multilingual NLP
An ML decision system that treats commodity purchasing as a sequential decision problem rather than just a forecasting problem.
Experiments with XGBoost, Temporal Fusion Transformer, and DQN to generate Buy / Split / Wait recommendations, exposed through a REST serving layer.
XGBoost TFT DQN Reinforcement Learning Time Series REST API
AI / ML
LLM APIs Tool Calling RAG Embeddings PyTorch scikit-learn Reinforcement Learning Whisper LangChain
Backend / Systems
Python FastAPI Flask Django Spring REST APIs SQL
Frontend
React TypeScript Vue.js
Cloud / Infrastructure
AWS Lambda API Gateway S3 GitHub Actions Render
Enterprise Systems
SAP Ariba BW/SAC CAP HANA
I'm interested in the space between a model that works in a notebook and an AI system that actually works.
That means thinking about more than model accuracy:
retrieval · evaluation · latency · grounding · failure modes · human feedback · system integration · deployment
I like experimenting with new AI techniques, but I'm even more interested in understanding when they work, when they fail, and how to turn them into reliable systems.