Operational intelligence for climate-resilient coastal critical infrastructure
Atlas4D is a decision-grade trust and evidence layer for ports, coastal municipalities, utilities, and public authorities operating under real-world uncertainty.
Atlas4D is a working deep-tech platform โ not a slide deck. A TRLย 6 validated prototype for climate-resilient coastal critical infrastructure.
Critical infrastructure operators already have data, but lack a system that determines in real time which signals remain reliable under degraded conditions. This creates operational blind spots exactly when decisions matter most.
Atlas4D is a decision-grade trust and evidence layer that separates observed reality from predictive output. It fuses heterogeneous signals and suppresses weak forecasts instead of presenting misleading certainty.
TRLย 6 validated prototype โ 45.1M+ observations processed, 680k+ trust metric points, and 4 signed Letters of Intent from regional pilot partners.
Ports, coastal municipalities, utilities, and public authorities operating in climate-stressed coastal environments.
Auditable AI, evidence lineage, uncertainty-aware decision support, and a deep spatiotemporal data foundation. Core differentiation: not more predictions, but more truthful operational state.
Multi-site operational pilots, deeper integration with operator workflows, and conversion of regional pilot traction into repeatable deployment across European coastal infrastructure.
EIC Accelerator 2026 applicant โ Proposal 101339041 ยท TRLย 6 ยท โฌ2.5M requested.
"Atlas4D doesn't log events-it logs the history of movement."
Static data
Temporal data
Predictive analytics
We work with history, not just "state". No overwrites-versioned intervals (valid_from โ valid_to).
STSQL: time + space + motion. Example: "Which drones approached the airport in the last 30 minutes?"
Trajectory analysis, vectors, and predicting positions N minutes ahead. Alerts before the event.
Dozens of specialized services-independent yet coordinated (IoT, AI, Weather, Vision).
Idempotency, Snowflake ID, dual-layer auth. Resilient to restarts and network loss.
Digital twin of reality with LLM/NLQ, vector trajectories and predictive models.
Atlas4D is a 4D spatiotemporal data layer โ a database-backed platform that mirrors the real world across space and time.
Instead of separate systems for GIS, time-series, events and logs, Atlas4D keeps movement in space and time in one model โ entities, trajectories, anomalies, threats and events.
Built on PostgreSQL + PostGIS + TimescaleDB + H3 + pgvector, Atlas4D behaves like a 4D RDBMS: SQL-compatible, observable and production-ready, but with native support for geospatial, temporal and vector AI.
Radar, weather, IoT, video, network telemetry and events are ingested into a single spatiotemporal model โ then exposed via maps, APIs, STSQL and NLQ for ports, coastal municipalities, utilities, and public authorities.
Atlas4D is designed as an infrastructure layer: it runs on your hardware, next to your sensors and existing systems, and complements your current monitoring, GIS and analytics stack instead of replacing it overnight.
Our spatiotemporal extension for real-time queries
-- Drones near the airport in the last 30 minutes
SELECT *
DURING ["now-30m","now"]
NEAR POINT(42.6977,23.3219)
WITHIN 5km
LIMIT 100;-- Predict position in 10 minutes and check a restricted zone
SELECT entity_id, predicted_position
FROM trajectories
WHERE PREDICT(+10m)
AND INTERSECTS(restricted_zone);Everything for spatiotemporal data analytics
PostGIS, H3 indexes and STSQL for massive real-time spatial data.
TimescaleDB continuous aggregates and materializations for fast dashboards and alerts.
MapLibre GL + MVT/PMTiles-millions of points, smooth render without blocking.
Integration with Ollama LLM: multilingual NLQ, STSQL generation, trajectory forecasts.
IoT, Vision, Radar, Weather, Fusionโฆ each service is a domain expert.
Anomaly detection and predictive alerts with fusion and ML.
โ Shipped: Vision Trust (Wilson CI, split freshness, worst-label). Weather truthish partial. MALP ฮณ=0.996.
Wilson CI 95% precision bounds, split freshness (recorder vs human loop), worst-label outliers, real-time UI badge.
โ Shipped: Scenario contexts (Reality/WIND_STORM). ๐งญ Roadmap: Monte Carlo what-if runner.
โ Shipped: valid/transaction in observations, issued_at in forecasts. ๐ก Expanding: as-of API across domains.
Atlas4D in action - operational intelligence for coastal critical infrastructure
Vehicle positioning with MVT tiles
MapLibre GL with Bulgaria coverage
Filters by time, type, severity and location
H3 hexagonal binning with real-time data
Specialized services working in sync
MapLibre GL + React interface
Routing, auth, rate limiting
JWT authentication
External data ingestion
IoT device processing
Camera feed integration
Natural Language โ STSQL
AI video/images
Vectorization
Forecasts
Pattern-based
Correlation
Predictive analysis
Realtime radar
Audio analytics
Orchestration
Merging
H3-indexed
Integrations
WebAssembly
ML inference
Legacy bridge
AI-powered risk assessment combining LSTM trajectory predictions, weather hazards, and anomaly patterns
Trajectory risk score
20 trajectories analyzed
Wind & storm risk
Real-time conditions
~0.1 kmยฒ cells
174m edge length
Prediction horizon
Configurable per request
Score: 0.0 - 0.3
Score: 0.3 - 0.7
Score: 0.7 - 1.0
POST /api/threats/forecast// Request { "lat": 42.5048, "lon": 27.4626, "horizon_minutes": 30 } // Response { "threat_level": "medium", "threat_score": 0.313, "trajectory_score": 0.897, "weather_score": 0.255, "confidence": 1.0 }
Real-time network infrastructure monitoring with geospatial intelligence and semantic search capabilities
โโ Network Devices โ โโ SNMP/Ping Workers โ โโ ping_worker (30s intervals) โ โโ snmp_worker (5min intervals) โ โโ Template-based collection โ โโ Data Layer โ โโ TimescaleDB (metrics hypertables) โ โโ PostGIS (geospatial geometries) โ โโ H3 (hexagonal indexing) โ โโ pgvector (embeddings storage) โ โโ Processing Layer โ โโ Ollama (embeddings generation) โ โโ Alert Worker (threshold monitoring) โ โโ STSQL Adapter (query compilation) โ โโ Presentation Layer โโ MVT Tiles (MapLibre GL) โโ REST API (FastAPI) โโ Prometheus Metrics โโ Web Dashboard
GET /api/netguard/search
// Query devices with performance issues near Burgas { "query": "network devices with high latency in Burgas region", "location": { "lat": 42.5048, "lon": 27.4626 }, "radius_km": 10, "time_range": "last_24h" } // Response with hybrid semantic + spatial + temporal results { "results": [ { "device_id": "lb.digicom.bg", "semantic_score": 0.89, "avg_latency_ms": 145.7, "distance_km": 2.3, "alert_count": 3 } ] }
AI-powered attendance prediction with Maximum Adjusted Linear Prediction (MALP) calibration for bias-corrected forecasts
1. Base Prediction (Heuristic Model) โโ Multi-factor analysis โ base_prediction: 4500 2. MALP Calibration (Bias Correction) โโ ฮณ coefficient: 0.9962 โโ Formula: calibrated = ฮณ ร base_prediction โโ Result: 4390 (adjustment: -110 or -2.4%) 3. Risk Assessment โโ Capacity utilization: 4390 / 5000 = 88% โโ Weather risk: precipitation probability โโ Traffic risk: real-time observations โโ Combined risk score: 0.3 (Medium) 4. Response with Transparency โโ predicted_attendance: 4390 โโ base_prediction: 4500 โโ calibration_method: "MALP" โโ adjustment: -110 โโ confidence: 0.65 Benefits: โ Reduced bias โ more accurate predictions โ Conservative estimates โ better planning โ Full transparency โ trustworthy results
Plan staffing, security, and resources with bias-corrected attendance forecasts
Optimize capacity, pricing, and logistics with MALP-calibrated predictions
Coordinate transport and emergency services with accurate crowd size estimates
Access full transparency: base vs calibrated predictions with adjustment deltas
POST /api/events/risk/assess
// Request - predict with MALP calibration { "event_name": "Summer Music Festival", "event_type": "concert", "venue_capacity": 5000, "location": { "city": "Burgas", "lat": 42.5048, "lon": 27.4626 }, "date": "2025-07-15T20:00:00Z" } // Response - MALP transparency included { "success": true, "predicted_attendance": 4390, "base_prediction": 4500, "calibration_method": "MALP", "adjustment_delta": -110, "adjustment_pct": -2.4, "risk_assessment": { "overall_risk": "ะกัะตะดะตะฝ", "risk_score": 0.3, "attendance_risk": 0.88, "weather_risk": 0.15 }, "confidence": 0.65 }
GET /malp/stats
// Response - calibrator status { "calibrated": true, "n_events": 10, "gamma": 0.9961950734747499, "ccc_before": 0.9855, "ccc_after": 0.9733, "ccc_improvement_pct": -1.24, "data_source": "production", "last_calibration": "2025-11-11T14:20:22" } // Training endpoint POST /malp/train?force=false // Single prediction calibration POST /malp/calibrate?prediction=7500
Short version of the Atlas4D investor story: problem โ whatโs built โ where weโre going.
Cities, telcos and infrastructure operators drown in spatiotemporal chaos: radar, weather, IoT, cameras, events, network monitoring โ all in separate systems, with no unified 4D view and no "brain" that can predict what comes next.
Decisions are made on Excel, screenshots and phone calls, not on a live model of reality.
Atlas4D is founder-built, bootstrapped and deployment-proven. Next step: productizing the 4D data layer for 1โ2 lighthouse customers.
Atlas4D has applied to EIC Accelerator 2026 (Proposal 101339041) and is open to selective equity conversations:
๐ Full investor brief: atlas4d.tech/investors-brief
The best for maximum performance
Recent development milestones from Sprint 1โ13
Context-aware follow-up queries. Ask "What's the weather in Burgas?" then "And in Sofia?" โ it remembers.
Redis intent cache, Ollama pre-warming, weather fast-path. From 21s to 70ms for common queries.
Unified header across all pages with hamburger menu. 7 navigation items: Dashboard, NLQ, Threats, STSQL, NetGuard, Events, Ops.
Optimized spatial scoring and threshold-aware processing. Real-time threat zone rendering at 0.1ms.
13k+ observations, SNMP/Ping workers, 11 ML models per device, MVT tiles, semantic search with pgvector.
AI attendance prediction with bias-corrected MALP calibration. Weather, traffic, and capacity risk scoring.
From safety to smart cities and industry
Drone incursions, anomalous trajectories, predictive alerts.
Storm early warning, flood risk, cross-municipal climate resilience coordination.
Load forecasts, coverage optimization.
Tracking, predictive maintenance, logistics.
Routes, collision avoidance, optimal trajectories.
Massive streams, anomalies, smart notifications.
Temporal monitoring and change analysis.
H3 grid, Open-Meteo integrations and correlations.
Early warning, spread prediction, evacuation planning.
Yield forecasting, irrigation optimization, pest detection.
Atlas4D is already running as a working deep-tech prototype. Looking for investors and strategic partners for the first large-scale deployments.