Decorative: a 3D grid city sits behind this page. Its towers are the 215 services Krish is on call for; an amber tower is an incident and a cyan pulse is the agent answering it. Everything shown in it is also in the page text.

KRISH ARYANon-call: green

Work / 02 · In production

A ReAct agent that reads logs the way an engineer would: with grep.

log lines / service / day
200k–30Mlog lines / service / day
ingestion pipelines
0ingestion pipelines
vector databases
0vector databases
01

Problem, approach, result

Problem

  • Each service writes 200k–30M log lines per day. Answering a question about them usually starts with building an ingestion pipeline.

Approach

  • A ReAct agent whose tools are bash: grep, awk and zcat over the raw logs.
  • No ingestion pipeline and no vector database.
  • A terminal UI (prompt_toolkit, Rich) plus a browser UI, on Azure AI Foundry.

Result

  • In production.
02

How it works

The agent writes the grep; the logs never move. Matches highlight as they stream past.

03

Stack

hover a tag to light everything that uses it