Home / Research / Labs

Research that’s almost in production.

A snapshot of LENS Labs - the work that lives between the demo and the rollout. Wireless sensing. Cross-modal scene reconstruction. Satellite intelligence. Molten-steel diagnostics. Some of these are in pilot. Some are in early customer hands. None of them are vapor.

Live · in pilot or with first customers In development · engineering active Early research · exploratory

Lab 01

LENS Pulse

Wireless sensing · presence, posture, action recognition from RF signals.

In development

Seeing through walls, without cameras.

Wi-Fi and ambient RF transmitters already light up every space we operate in. LENS Pulse reads channel state information (CSI) from off-the-shelf access points and infers presence, count, posture and action - through walls, in the dark, with no privacy-invasive optics. Built first for defence and critical-infrastructure customers where camera placement is impossible or undesirable.

Signal

802.11n/ac/ax CSI · sub-6 GHz RF

What we infer

Presence, count, posture, gait, fall-detection, action class

Stage

Engineering active · partner-led pilots

Defence researchBorder securityEldercareIndustrial safety

Lab 02

Cross-modal 3D

Real-time 2D CCTV -> 3D scene reconstruction · analyst-ready volumetric view.

In development

From flat feeds to volumetric scenes.

Conventional CCTV walls leave the cognitive lift to the operator. We reconstruct a unified 3D scene from multi-camera 2D feeds in real time - people, vehicles and assets as tracked entities in a shared spatial frame. Operators get one pannable, zoomable scene instead of forty rectangles. Built on the same edge substrate as LENS View; runs on commodity GPUs.

Inputs

RTSP/ONVIF camera arrays · calibrated or self-calibrating

Output

Real-time volumetric scene · query-able by entity

Stage

Pilot with municipal command centre

Smart citiesStadium & transitIndustrial floorsCritical infrastructure

Lab 03

NIST ELFT

Latent-fingerprint biometrics research · benchmark-led model improvement.

Live

Forensic-grade fingerprint matching.

NIST's Evaluation of Latent Fingerprint Technologies is the canonical benchmark for forensic biometrics. We run our matchers against it on every release - partial prints, smudged prints, prints lifted from real crime-scene surfaces. Results feed directly into LENS Identity, used by national police forces and Project Mukti for rapid-recovery scenarios involving infants and unidentified persons.

Benchmark

NIST ELFT-EFS · PFT-III

Templates

Hardened, FIPS-grade, on-device matchable

Stage

In production via LENS Identity

Police & forensicsProject MuktiKIET infant biometricsNational ID

Lab 04

Satellite intelligence

Inference from satellite-borne signals, video, audio and noise.

Early research

From orbit to actionable.

Satellite constellations now downlink signal, imagery, audio and broadband noise at rates that exceed any analyst pool. We're building inference pipelines that turn raw orbital data into measurable, actionable events - vessel-tracking from RF emissions, infrastructure-change detection from imagery, anomaly classification from broadband noise. Same intelligence substrate, different sensor frame.

Sensors

Optical, SAR, RF, broadband audio

Inference

Vessel tracking, infrastructure change, anomaly classification

Stage

Architecture & partner discussions

Maritime domain awarenessDisaster responseEnergy sectorDefence & strategic

Lab 05

Molten-steel diagnostics

Defect & impurity detection during steel production, at the molten stage.

Live

Catching defects at 1,500 °C.

Conventional steel quality control catches defects after the slab cools - hours after the impurity entered the melt. Working with ArcelorMittal, we've trained vision models to detect inclusion patterns and chemical anomalies during the molten stage, when corrective intervention is still cheap. The same pipeline now runs on JSW's MIDAS furnace floor, classifying scrap charge and predicting heat outcomes.

Conditions

Furnace cameras · 1,400-1,650 °C operating

Detected

Inclusion patterns, slag layer, chemical anomalies

Stage

In production at ArcelorMittal & JSW

ArcelorMittalJSW Steel · MIDASFoundry analytics

Lab 06

LENS LM

Domain-adapted language models for command, dispatch and operator co-pilot.

In development

An operator co-pilot, not a chatbot.

Off-the-shelf LLMs hallucinate when the stakes are operational. LENS LM is fine-tuned on dispatch, incident-command, public-safety and industrial-control corpora - with hard guardrails, retrieval grounding and explainability surfaces. Integrated as the natural-language layer across LENS View, Industry and Vehicle.

Base

Open-weights LLM · LoRA-adapted on domain corpora

Mode

RAG-grounded · cited responses · explainable refusals

Stage

In production for selected operator UIs

Command centresIndustrial floorsHealthcare ops

Partner with LENS Labs

Most of what ships at LENS started here.

Crowd-density forecasting was a research line in 2019 - it’s now LENS SAFER, deployed at Mahakumbh. Latent-print matching was a benchmark exercise - it’s now in police case-management workflows. If something on this page maps to a problem you can’t solve in production today, talk to us.

Bring us a problem →