Problems / Agents and tools, medium
An agent service samples 5% of its trajectories, 12 a second, into a stream that keeps 20 seconds of data. LLM-judge consumers score them for regressions, but four consumers manage only about 7 a second.
Unlock it once for 15 credits; everything inside it is free (runs, changes, Submit, results, hints). Sign in with Google to start; new accounts get free credits.
No trajectories lost to retention; success ≥ 99.9%.
Like a coding problem, a solution has to pass every case, not just the one it was built for. Submit in the lab runs all 5; 2 of them are hidden until then.
12 requests a second for 60 seconds: 100% trajectory (writes).
What each component does in this design, and what calls it. The simulator mimics each one from these settings.
| Component | Its job here | Settings |
|---|---|---|
| Load balancerLoad balancer | Traffic enters here.Calls Agent service (5% sampled).Layer 4 balancers route by IP and port; layer 7 balancers read HTTP and can route by path or header. | – |
| Agent service (5% sampled)App service | Load balancer forwards to it.Calls Trajectory stream (publishes).Little's law: throughput ≈ concurrency ÷ latency. 200 slots at 50 ms serve about 4,000 req/s. | 3 replicas × 500 slots · timeout 1.0 s |
| Trajectory streamEvent stream | Agent service (5% sampled) publishes to it.Calls LLM-judge consumers.Kafka keeps order only within a partition, so the partition key decides what stays in sequence. | 12 partitions · retention 20 s |
| LLM-judge consumersWorker | Trajectory stream consumes from it.Queues deliver at least once, so a worker can see the same message twice. | 4 concurrent jobs |