Build a Hosted Claude Client-Side Latency and Budget Observability Engine

Track client-side HTTP request latencies and per-call token costs for hosted Claude API workloads. Flag budget pacing anomalies and latency threshold breaches for platform engineering oversight.

AI Platform Reliability Engineer

Ingest normalized client request telemetry logs containing round-trip time and token usage.

Calculate token consumption costs against published tier pricing and verify budget burndown velocity.

Detect client-side p95 latency degradations and error bursts across caller application IDs.

Generate Claude-assisted platform incident briefings with recommended client retry pacing limits.

Web dashboard displaying real-time token spend gauges, latency percentiles, and downloadable incident reviews.

Construct a synthetic log stream of 500 Anthropic API client responses with client timestamps and token counts. Starter examples below are synthetic. They are not downloaded records.

Anthropic Python SDK Utilize Anthropic Python SDK client invocation standards, message structures, and usage counter patterns.

langfuse, anthropic, pydantic. langfuse provides open tracing abstractions to capture client-side duration and token metrics cleanly. Hosted Claude API accessed over HTTPS for log explanation and diagnostic report generation.

Submit runnable source, pinned dependencies, setup commands, a working interface, example outputs, and a recorded end-to-end demo. Include tests and an action log.

Computes cumulative input and output token spend strictly from client usage metadata.

Flags client application endpoints exceeding configured latency SLAs without assuming server internals.

Synthesizes structured incident action items for client-side connection pool tuning.

```json [ { "input": { "app_id": "customer-support-bot", "requests": 150, "total_input_tokens": 120000, "total_output_tokens": 30000, "client_p95_ms": 4200, "daily_budget_usd": 50 }, "expected_output": { "calculated_spend_usd": 0.81, "budget_pacing": "healthy", "latency_sla_breach": true, "recommendation": "enable_streaming_or_reduce_max_tokens" } }, { "input": { "app_id": "batch-document-indexer", "requests": 500, "total_input_tokens": 25000000, "total_output_tokens": 4000000, "client_p95_ms": 11500, "daily_budget_usd": 100 }, "expected_output": { "calculated_spend_usd": 135, "budget_pacing": "exceeded", "latency_sla_breach": true, "recommendation": "throttle_client_concurrency_apply_batch_api" } } ] ```

Performs client-side telemetry observability only, making no assertions about proprietary model server hardware or internal routing. Use mock external writes. Require human review before any real deployment.