Build a Remote Telemetry Anomaly Agent with Vercel AI SDK and Gemini 2.5 Flash
Industrial remote monitoring systems struggle to isolate real-time sensor anomalies from transient operational noise. Build a streaming telemetry analysis agent using Vercel AI SDK and Gemini 2.5 Flash that ingests OPC UA time-series data and flags true anomalies. Achieve a 95% anomaly detection accuracy while maintaining under 200ms processing latency per batch.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop a real-time SCADA telemetry anomaly agent using Vercel AI SDK streaming pipelines connected to Gemini 2.5 Flash model tooling.
How work is evaluated
Evaluates streaming anomaly detection precision, recall, and end-to-end processing latency on synthesized OPC UA stream data.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
latency_check
Ensure telemetry batch processing completes in under 200ms
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
anomaly_precision
Ratio of true positive anomalies detected • target: 0.95 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Stream OPC UA telemetry directly into an AI SDK streaming pipeline
Configure Gemini 2.5 Flash with custom function tools for time-series thresholding
Evaluate anomaly precision and false-alarm reduction on live factory telemetry
Reference links and supporting material
Synthetic OPC UA time-series dataset simulating motor vibration, bearing temperature, and pressure fluctuations across normal and faulted operation states.
Challenge setup
Evaluates streaming anomaly detection precision, recall, and end-to-end processing latency on synthesized OPC UA stream data.
Challenge input
JSON array of timestamped sensor readings (vibration, temp, pressure)
Vercel AI SDK
Provides fast streaming tools and multi-provider TypeScript bindings.
Gemini 2.5 Flash
Fast sub-second reasoning and function tool execution for high throughput.
Evaluated output
JSON object with classified anomaly alerts and confidence scores
- Ensure telemetry batch processing completes in under 200ms
- Ratio of true positive anomalies detected • target: 0.95 • range: 0-1
- Benchmark: Industrial Sensor Anomaly Suite (ISAS-2025)
- Anomaly Precision target: 0.95
- 1 public reference case
- JavaScript execution harness
- javascript sandbox (unavailable on Versalist)
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- Vercel AI SDK · Required
- Google Gemini · Optional
- Gemini 2.5 Flash · Optional
- Vercel AI SDK · Required
Evaluation contract
- latency_check · Weight 1
- anomaly_precision · Weight 1
Metrics
- anomaly_precision · Target 0.95
Benchmarks
- Industrial Sensor Anomaly Suite (ISAS-2025)
- energy
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON array of timestamped sensor readings (vibration, temp, pressure)
Output: JSON object with classified anomaly alerts and confidence scores
Execution harness: JavaScript
Evaluation tools
- Vercel AI SDK · Agent Orchestration Framework
- Gemini 2.5 Flash · Inference LLM
Recipe freeze marker
No recipe freeze time is recorded.