• 🤖 Learning-based Planning: Research on scalable multi-goal path planning and heuristic priors for classical planners ($A^$, RRT).
• 🧠 Agentic AI: Designing autonomous exploration Agents with custom control flow — Phase-based state machines, MCP tool protocols, and auditable reasoning chains. Building a multi-Agent system where Graph Exploration serves as a heavy reasoning tool.

• fin-trace — Graph exploration Agent with MCP. Custom Phase state machine (EXPLORING → FINALIZE) replaces standard ReAct. Event Thread reasoning with ku_id evidence anchors for end-to-end auditability. Deployed as standalone TS process, integrable by other Agent systems.
• fin-graph — Financial Knowledge Graph RAG pipeline. Entity-Event bipartite graph with 10K+ entities and 20K+ events. Dual-recall (BM25 + FAISS) → Rerank → graph expansion → MCP service.

Deep Learning & Algorithms

Agentic AI & Backend