Getamazednow AI · GenAI Project Observability
Demo dashboard · synthetic data · Order Support & Returns Assistant
MOCK / SYNTHETIC DATA
This dashboard renders entirely synthetic, generator-produced data (data/generator/) to demonstrate an observability approach for agentic Gen AI workflows. It is not connected to any production system. See README for scope, and /datadog for how this maps to a real Datadog implementation.
Daily workflow volume by outcome
Success rate (%) — daily trend
Cost per successful workflow (USD) — daily trend
Workflow volume by tenant (brand)
Workflow volume by use case
Cost composition — 30-day total by category
Incident log (Sev1/Sev2)
Latency by layer (P95, ms) — daily trend
End-to-end latency — P50 / P95 / P99 (ms)
LLM error rate, retry rate & tool error rate (%)
Provider rate-limit events — daily
Agent loop events — daily
Representative trace waterfall — single workflow, incident day (rate-limit + fallback)

Illustrative single-trace breakdown showing how a workflow's end-to-end latency decomposes into guardrail, retrieval, model (with a rate-limited retry + provider fallback) and tool steps — the trace-tree view a real APM/LLM-observability tool would render per request.

Prompt injection attempts — blocked vs bypassed (daily)
Sensitive data egress & toxicity flags — daily
High-risk tool calls & policy approval bypasses — daily
Incident spotlight — INC-2026-0611 (prompt-injection policy bypass)
Token usage — input vs output (daily)
Cost by model (daily, USD)
Provider quota utilisation (%) — daily

Dashed line marks the daily LLM-call quota assumption. The promo-driven rate-limit incident (days 21-22) is exactly the kind of quota breach this panel is designed to catch before it becomes an outage.

Cumulative cost vs. monthly budget
Fallback-model cost uplift (daily, USD)

Cost incurred specifically by routing to the (more expensive) fallback model during provider degradation — the direct dollar cost of the rate-limit incident.

Average step count per workflow — daily
Average tool calls per workflow — daily
Tool calls by name (30-day total)
Human escalation rate (%) — daily
Tool call success rate (%) — proxy for tool-selection quality

Proxy metric: this demo doesn't model ground-truth "correct tool for intent" labelling, so tool-call success rate stands in for tool-selection accuracy. A real evaluation harness would score selection directly against labelled intents.

Retrieval hit rate (%) — daily
Groundedness & citation accuracy — daily average
Hallucination rate (%) — daily
Abstention rate (%) — daily
Average retrieved-source freshness (days) — daily
Regression pass rate & golden-set accuracy (%) — daily

Sampled from a synthetic evaluation harness (golden-set + regression suite), run on a schedule separate from live production traffic — not derived from raw request spans, matching how a real eval pipeline operates.

Release / rollback log