Building The Impossible.

Aviation securityAPID · Automated Prohibited Item Detection

Trained on the physics of X-ray, not just pixels.

Most screening AI learns what a threat looks like. APID also learns why it looks that way: how photons are absorbed through stacked materials. That lets it separate organics, metals and dense clutter on physical grounds and flag what a pixel-only model would miss.

Decisions on every bag

2

AI maker, human checker

Model heads

6

Each with its own confidence threshold

Balanced dataset

3 lakh

Images generated on demand

Analyst views

6

HI-MAT to invert and B/W

How the models learn

1. Attenuation modellingBeer–Lambert forward models generate physically valid scans across densities, thicknesses and overlaps.
2. Material discriminationDual-energy decomposition gives every pixel an effective atomic number and areal density.
3. Synthetic threat projectionRare threats are projected physically into real bag scans, so models train on balanced data without staging live threats.
4. Multi-view fusionDetections are reconciled across views to resolve occlusion and suppress single-view artefacts.

Operating model

1. MakerAI ensemble flags class, region and material signature in real time
2. CheckerThe screener confirms or overrides each call; disagreement escalates
3. AuditEvery decision is logged and fed back into retraining

Dataset factory

• Balanced · 3 lakh
• Threat-heavy · 1 lakh
• Concealment · 50k
• Suitcase focus · 50k
• Handbag & purse · 25k
• Clean negatives · 10k
• One-click YOLO export

• Probability of detection
• False alarm rate
• Bags per hour
• Operator decision time
• Custom sensing fusion

Spatial computingRenewable-park digital twin

One twin, from the whole park down to a single module.

Every project and asset is mirrored in a twin that combines the schedule network, cost model and generation model, with every physical element bound to the ontology. Cameras are registered in 3D, weather is live, and robots and drones are planned inside the same scene they will work in.

Zoom levels

4

Portfolio, block, table, module

Weather states

4

Clear, dust, overcast, fog

Camera types

Real & virtual

Plan coverage before installing

Every module

ID'd

Specification and live status

What the twin does

1. Portfolio to modulePlots and capacity across the park, block boundaries and tracker rows, every table addressable, each module with its ID and status.
2. Soiling, module by modulePer-module soiling from turbine-mounted and drone cameras ranks cleaning by generation loss, then becomes work orders.
3. Field operationsCleaning robots simulated row by row; drone missions drawn as waypoints on the block map, then flown with live telemetry.
4. Any weather, any hourTurbines respond to live wind speed and direction across clear, dust, overcast and fog.
5. Cameras in 3DEach camera's field of view is modelled on site geometry, so a detection becomes an event at a known place, linked to the asset and the work order.

• Construction-progress twins
• Urban infrastructure
• Design validation
• Predictive simulation

AI workflow engineCortex

From a camera feed to a governed decision, configured rather than coded.

Cortex binds every signal to the asset it describes, runs it through versioned models and rules, and ends in an alert, a work order or a dashboard. A new use case is a new graph, not a new software project.

Layers

6

From ingestion to governed action

Models

50+

Vision, time-series, anomaly and LLM, and growing

Workflow to app

1 click

Export any graph as a dashboard

Insights

Private

An on-premise LLM writes them; data stays on site

Six layers, raw stream to governed action

1. Ingestion fabricImages, RTSP video, drone and satellite imagery, IoT telemetry and enterprise events on one time base.
2. OntologyPlants, blocks, assets, work orders and vendors, so every signal binds to its asset.
3. Workflow graphA versioned graph of typed operators: ingest, detect, enrich, reason, decide, act.
4. Model registryVision, time-series and anomaly models with drift monitoring and A/B evaluation.
5. Reasoning & decisionRules, thresholds and LLM agents add severity, root cause and next action.
6. Action & governanceAlerts and work orders behind approval gates, with role-based access and a full audit trail.

• PPE
• Driver fatigue
• Hairnet & glove
• Hand-wash
• Road damage
• Lane detection
• Fire & smoke
• Weapons
• X-ray threats
• Vehicle damage
• Licence plates
• Animal & bird species
• …and more

Decision intelligenceEnterprise Intelligence Platform

Every project, line and alert, on one live screen.

A renewable portfolio runs on schedules, ERP, procurement, daily progress reports, transmission readiness and SCADA. The platform lands all of it in one layered lakehouse, links it into a graph, watches it with risk engines, and prices the options in a scenario lab before leadership decides, then writes the chosen action back to the systems of record.

Source families

7

Unified, from schedules to SCADA

Risk engines

4

Schedule, supply, evacuation, operations

Role cockpits

4

Leadership to techno-commercial

Before every decision

Monte Carlo

Date, cost and generation impact

What the platform shows

1. Portfolio commandStatus, progress against plan, CAPEX, procurement, MW and lines charged, with every transmission corridor on the map.
2. Actual against expectedA portfolio S-curve beside every project's completion and schedule-variance index.
3. Every daily progress report, readManpower, days lost, why work stopped and who needs attention, rolled up automatically.
4. Risk consoleEvery project scored 0–100 and ranked, with the drivers one click away.
5. Grid connectivityPooling station to substation to project to block, with MW roll-ups from the transmission portal.

Detect to execute, in one loop

1. DetectA risk crosses its threshold
2. GenerateRe-sequence, crash, switch vendor
3. SimulateMonte Carlo on each option
4. DecideLeadership compares options
5. ExecuteAgents draft changes behind approval gates
6. LearnModels recalibrate on outcomes

• Primavera P6
• SAP
• Ariba
• Daily progress reports
• Transmission
• SCADA & O&M
• Field sensors & drones

Battery storageUnada EnergyOS

The battery fleet, live: every site, container and rack.

An AI-powered digital twin and operations command centre for battery energy storage. It follows energy from generation through storage to the grid, and on to green hydrogen and ammonia, from the control room to the boardroom.

Sites in the demo fleet

10

Across India

Fleet modelled

4.0 GWh

2 GW peak

Telemetry

Rack‑level

Health and remaining useful life

Domain monitors

7

Solar, wind, hydro, grid, finance, vendors, DPRs

From the fleet to a single rack

1. Command centreCapacity, stored energy, health, alarms, revenue, CO₂ avoided and uptime, with a live energy topology you can click into.
2. Site twinEvery container in 3D, coloured by temperature from cool to critical, with alarms flagged where they sit.
3. Container & rackState of charge and health, temperature, voltage, current and power, trends and remaining useful life for all twenty racks.
4. Thermal & predictive AIHotspots and cooling efficiency, then failure probability and days-to-failure per component, with root cause, action and cost.
5. Copilot & simulatorAn AI engineer that explains root cause, risk and financial impact, and a what-if lab for heat, load, cooling and grid stress.
6. Electron to moleculeSolar and wind through storage and electrolysis to hydrogen, ammonia and the grid, in one value-chain flow.

• Alarm centre
• Asset analytics
• Executive view
• Green hydrogen
• India twin map

PerceptionVisual understanding & multi-sensor fusion

Every sensor on site, on one clock.

A camera event corroborated by a SCADA alarm, or a placement on a smart surface that confirms what a camera saw, is far more trustworthy than either alone. Our fusion engine aligns every signal in time and space, then confirms across modalities.

Camera types

5

Fixed, PTZ, drone, body-worn, thermal

Protocol families

3

OPC-UA, Modbus, IEC

Time base

1 clock

Every signal aligned

Smart surface

24

Electrodes, read about 50× a second

Beyond off-the-shelf detection

1. Multi-camera trackingRe-identification across cameras.
2. Instance segmentationPixel-accurate objects, not just boxes.
3. Pose & actionUnsafe acts and behaviour from motion.
4. 3D reconstructionGeometry recovered from imagery.
5. Change detectionWhat moved, grew or degraded over time.
6. Edge & active learningOptimised for site hardware; corrections become training data.

Four applications on one perception base

1. Construction monitoringEquipment, materials, manpower and safety compliance across sites in real time.
2. Progress monitoringDrone and camera imagery against plan and BIM to quantify progress per activity.
3. Behavioural detectionUnsafe acts, intrusion, loitering, PPE violations and abnormal crowd behaviour.
4. Pattern detectionRecurring anomalies across time and sites, linked to sensor trends to surface root causes.

• Smart shelves & tool cribs
• Tamper & removal alerts
• Interactive experience tables
• Confirming CCTV events