1import time
2import uuid
3from contextlib import contextmanager
4
5@contextmanager
6def llm_span(name: str, **attrs: object):
7    trace_id = str(uuid.uuid4())
8    started = time.perf_counter()
9    try:
10        yield trace_id
11    finally:
12        elapsed_ms = int((time.perf_counter() - started) * 1000)
13        event = {"trace_id": trace_id, "name": name, "latency_ms": elapsed_ms, **attrs}
14        print(event)
15
16with llm_span("Cost Tracking", model="gpt-4o-mini", prompt_tokens=128):
17    response = "Grounded answer with cited context."
18print(response)