Multi-agent orchestration
Designing agent pipelines that coordinate inbox parsing, evidence extraction, state updates, and follow-up planning while keeping important mutations behind review gates.
AI Engineering
Multi-Agent Systems and AI Reliability
I build AI systems as software systems: agents with tools, retrieval, memory, evaluation, observability, and controlled state changes. The goal is not a simple chatbot demo; it is turning messy context into reliable workflows that can be inspected, measured, and improved.
Engineering Focus
Retrieval, memory, and evaluation
Building systems around citations, durable memory, failure cases, retrieval quality, and eval contracts instead of treating a model call as the product.
AI system optimization
Optimizing latency, cost, tool-call budgets, prompt/data contracts, model-serving constraints, and observability so AI features can survive real workflows.
Relevant AI Projects
Other Candidate
A Gemma-powered recruiting mailbox pipeline with Gmail evidence, career memory, multi-agent extraction, traceable review gates, and controlled application-state updates.
Public CareerOS Judge Demo
A local-first public proof package with sanitized recruiter threads, optional Gemma/Ollama analysis, deterministic fallback, model traces, and pipeline eval results.
cc-lite Xiangqi Research Baseline
A compact AlphaZero-style research baseline with Xiangqi rules, PyTorch policy-value modeling, legal policy masking, MCTS, self-play, training, and evaluation scripts.
Tripemini AI Travel Planner
An AI-assisted travel planning app that turns trip context into structured itinerary recommendations and product-ready planning flows.
Explore More
For broader engineering work, review the full project list or the cloud and data systems page. For opportunities, contact Hanbin directly.