THE PLATFORM
An escalation scoring engine, not a mentions dashboard.
Velantir ingests social posts for the entities you protect, scores escalation risk every five minutes through a three-layer model, and fires an alert only when threshold, persistence and confidence gates break together — with the evidence, the dominant trigger factor and a recommended action attached.
ARCHITECTURE
Ingest. Score. Alert. One five-minute loop.
INGEST
Hybrid discovery across five active platforms
Multi-provider ingestion routed per platform: Bright Data for TikTok and YouTube, Apify for X, Facebook and Instagram. Threads is supported but inactive by default. Every raw post is archived permanently as the source of truth for reprocessing.
- Keyword, hashtag and seed-account discovery legs
- Bahasa Indonesia and English classification
- Per-platform provider routing and capability registry
- Permanent raw archive for replay and recalibration
SCORE
Three layers, calibrated per industry
Signals are normalised, combined through interaction terms rather than a linear sum, then tested for persistence over time. Interaction weights and threshold multipliers are calibrated per industry model against historical escalation events before go-live.
- Layer 1 — velocity, amplification, convergence, narrative risk
- Layer 2 — Critical Interaction Score (CIS)
- Layer 3 — temporal persistence across rising windows
- Final — confidence-adjusted score on a 0–100 scale
ALERT
Gated alerts with the reasoning attached
No alert fires on a single spike. When all three gates break simultaneously and no cooldown is active, the alert engine assembles the full payload and delivers it to your channels within the same cycle.
- Dynamic per-entity threshold, not a global constant
- Narrative summary, evidence posts and key changes
- Predicted escalation and recommended action
- Feedback loop: relevant, not relevant, missed severity
CAPABILITIES
Built for the way risk and comms teams actually investigate.
Bilingual narrative classification
Every post classified into one of seven categories from Neutral through Boycott Call, with a model confidence value.
Amplification anomaly detection
Author engagement-to-follower ratios scored against the entity's own historical baseline distribution.
Dynamic per-entity thresholds
Each entity's alert threshold is its own baseline mean plus a calibrated multiple of its standard deviation.
Narrative clustering and phases
Embeddings indexed for similarity search, with phase transitions tracked and flagged when severity escalates.
Replay and reprocessing
Re-run any past window through a new model version against the raw archive — replay results never trigger alerts.
Six investigation modules
Escalation timeline, velocity curve, platform breakdown, cluster map, historical comparison and alert feed.
SECURITY & COMPLIANCE
Built to the standard enterprise security requires.
Velantir runs as a multi-tenant platform in the Asia Pacific region, with every query scoped to the client that owns the data and every write recorded in an audit log.
- Strict per-client data isolation on every query
- Role-based access: admin, analyst, read-only
- Managed identity with token-based API authentication
- Audit log on every create, update and delete
DEVELOPER API
Every signal accessible via API.
Every dashboard module is backed by a documented REST endpoint. Read escalation scores, velocity, platform breakdown, clusters and alerts, and submit alert feedback from your own systems.
GET STARTED
See the scoring engine run against your own entity.
A session with our engineering team covering entity configuration, keyword and seed-account setup, industry model calibration and alert routing.