Agent Engineering
Shows planner, tool routing, tool trace, evidence ledger, deterministic fallback, and local LLM compatibility.
Explainable Bond Analysis Agent
BondLens AI
一个从本科毕设升级而来的中文债券分析智能体项目，重点展示工具调用、实时数据兜底、证据账本、答案评审、风险画像、LLM guardrail、红队评测和 Docker/CI 工程化能力。
Core Workflow
BondLens AI keeps the financial reasoning path visible and auditable.
AkShare live data -> cached live snapshot -> local Excel fallback
Rule-based planner chooses market, search, ranking, outlier, and report tools
Python tools produce structured evidence before any optional LLM call
LLM numeric and risk-language guardrails decide whether LLM text can be final
Evidence ledger, answer judge, risk profile, and replay dashboard make runs reviewable
Agent evals and red-team evals run in CI without API keys
Why It Matters For Interviews
Financial Data Boundary
Clearly separates live data, cached snapshots, static samples, and missing issuer/credit/macro context.
Safety Evaluation
Blocks unsupported numbers, investment advice language, return guarantees, and overconfident risk claims.