beixu@home:~$ cat /work/agentic-systems.md

agentic systems

Evidence-grounded agents, mono-agent runtimes, clinical agents, and human-in-the-loop orchestration.

What this is

A line of work around agents that do more than chat: they decompose tasks, call tools, collect evidence, expose intermediate state, and leave a trail a human can review.

EviAgent

Evidence-based decision agent that turns a question into claims, cited evidence, an evidence graph, a decision, and verification steps.

Mono / mono-agent

Runtime experiments for controllable tool use, task execution, context handling, and human-in-the-loop operation.

ClinMono Agent

Clinical agent platform with LLM routes, screening workflows, Agent Lab surfaces, and research MCP integrations.

Agentic Ordering Service

Production-style multi-agent pipeline with intent extraction, planning, policy validation, and tool execution.

Design pattern

claims and tasks are separated from evidence collection
tool execution is explicit, inspectable, and policy-gated where needed
clinical and high-stakes flows keep human review in the loop
outputs include verification steps instead of pretending to be final truth