InertialAI-Embed-Alpha
A unified embedding model that encodes text and time-series into a shared semantic space—enabling cross-modal search and retrieval.
Cross-Modal Search
Search time-series patterns using natural language. InertialAI-Embed-Alpha is the first production-ready embedding model that unifies text and time-series in a shared semantic space.
Built on InertialAI-Alpha and fine-tuned with contrastive learning, it enables applications impossible with single-modality approaches—find sensor anomalies with queries like "sudden temperature spike followed by system failure" or cluster patient data by clinical descriptions.
Specifications
Matryoshka Embeddings
Trade dimensionality for efficiency. Smaller dimensions retain high quality.
Benchmark Results
Competitive with specialized models while uniquely supporting cross-modal retrieval.
Text Similarity (MTEB)
| Task | Ours | MiniLM |
|---|---|---|
| STS Average | 0.759 | 0.790 |
| ArguAna | 0.543 | 0.502 |
| SciFact | 0.668 | 0.645 |
Time-Series Retrieval (P@1)
| Benchmark | Ours | CHRONOS |
|---|---|---|
| UCR (Univariate) | 0.855 | 0.856 |
| UEA (Multivariate) | 0.625 | 0.601 |
Cross-Modal Retrieval (TimeMMD)
| Task | P@1 | P@5 | Random |
|---|---|---|---|
| Text → Time-Series | 0.535 | 0.915 | 0.431 |
| Time-Series → Text | 0.519 | 0.900 | 0.431 |
24% above random baseline. The only model enabling this capability.
Use Cases
Healthcare Monitoring
Search patient monitoring data using clinical descriptions. Query "irregular heartbeat preceding cardiac event" to retrieve similar ECG patterns. Achieves 0.785 P@1 on epilepsy detection.
Financial Analysis
Retrieve market patterns matching analyst commentary. Find historical periods described as "rapid drawdown followed by consolidation."
Unified Observability
Query logs and metrics in a single search. Find "database connection pool exhaustion" across log lines and metric spikes. Cross-modal P@1: 0.535.
Industrial Monitoring
Search sensor data using maintenance reports. Query "gradual temperature increase with vibration anomaly" across factory equipment.